Classification prediction model training method, application method, device and equipment
By deriveing and filtering the original variable data, and building and training a classification prediction model, the existing information push methods are solved, and more accurate and efficient information push is achieved.
Patent Information
- Application Number
- CN202310664550.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-06
- Publication Date
- 2024-12-06
AI Technical Summary
Existing information push or information delivery methods generally adopt the rough "wide net" model, resulting in high push costs, low conversion rates, and relying on work experience, low efficiency, and consuming a lot of labor and time costs.
By performing feature derivation, preprocessing and filtering of original variable data, a classified prediction model is constructed and trained, and the model is optimized to predict information push objects more accurately and improve information push efficiency.
Through feature optimization and model improvement calculation, more accurate information push target recognition is achieved, reducing push cost and time cost, and improving information push efficiency.
Smart Images

Figure CN119089191A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of artificial intelligence technology, and in particular to a training method, application method, device and equipment for a classification prediction model. Background Art
[0002] With the rapid development of artificial intelligence technology, many product owners will choose information push or information delivery methods so that the browsing objects can receive these push information, thereby achieving a promotional effect.
[0003] The current method of pushing or delivering product information is usually a rough and non-targeted approach, that is, a "wide-net" model, in which push information is delivered to all product trading objects at the same time in order to reach all product trading objects; or, based on the work experience of relevant staff, a part of product trading objects that may have transaction intentions are screened, and push information is delivered to these product trading objects in order to reach some product trading objects.
[0004] However, information delivery or information push generally generates conversion rate by counting the number of clicks on product push information, while delivering push information to all product trading objects at the same time through a "wide-net" model not only has high push costs, but also easily leads to low conversion rates due to the low stability of the number of clicks. In addition, based on the work experience of relevant staff, the method of reaching some product trading objects relies too much on the experience of the staff, is easily affected by the experience of the staff, and requires a lot of manpower and time costs, resulting in low push or delivery efficiency. Summary of the invention
[0005] The embodiments of the present application provide a training method, application method, device and equipment for a classification prediction model, which is used to better obtain feature data suitable for classification prediction model training by performing feature derivation operations, feature preprocessing operations and feature screening operations on original variable data, so as to better optimize the classification prediction model. Further, test feature data corresponding to the type of target feature data can be obtained to calculate the improvement of the classification prediction model, so as to more accurately obtain a target classification prediction model that meets business requirements, so that the information push object can be better predicted based on the target classification prediction model for information push, so as to improve the efficiency of information push.
[0006] On the one hand, an embodiment of the present application provides a training method for a classification prediction model, comprising:
[0007] Obtaining original sample data carrying label information, wherein the original sample data includes model training data and model test data;
[0008] Get the original variable data corresponding to the model training data;
[0009] Perform feature derivation operation on original variable data to obtain derived feature data;
[0010] Performing a preprocessing operation on the derived feature data to obtain preprocessed feature data;
[0011] Performing feature screening operation on the preprocessed feature data to obtain target feature data;
[0012] Constructing N original classification prediction models according to the model type, and performing model training operations on the original classification prediction models based on the target feature data and the label information corresponding to the target feature data, to obtain N trained classification prediction models, where N is an integer greater than 1;
[0013] Acquire test feature data corresponding to the type of target feature data in the model test data, and input the test feature data into N trained classification prediction models respectively to obtain N groups of test result data;
[0014] Perform lift calculation operations on N groups of test result data respectively to obtain test lift data corresponding to each group of test result data;
[0015] The trained classification prediction model with the highest test lift data is used as the target classification prediction model.
[0016] On the one hand, an embodiment of the present application provides an application method of a classification prediction model, including:
[0017] Input the user information to be predicted into the above classification prediction model, and output the prediction probability value through the classification prediction model;
[0018] If the predicted probability value is greater than the preset probability threshold, the product information is pushed to the predicted user;
[0019] If the predicted probability value is less than or equal to the preset probability threshold, the product information push operation will not be performed.
[0020] On the other hand, the present application provides a training device for a classification prediction model, comprising:
[0021] An acquisition unit, used to acquire original sample data carrying label information, wherein the original sample data includes model training data and model test data;
[0022] The acquisition unit is also used to acquire original variable data corresponding to the model training data;
[0023] A processing unit, used for performing a feature derivation operation on the original variable data to obtain derived feature data;
[0024] The processing unit is further used to perform a preprocessing operation on the derived feature data to obtain preprocessed feature data;
[0025] The processing unit is further used to perform a feature screening operation on the pre-processed feature data to obtain target feature data;
[0026] The processing unit is further used to construct N original classification prediction models according to the model type, and perform model training operations on the original classification prediction models based on the target feature data and the label information corresponding to the target feature data, so as to obtain N trained classification prediction models, where N is an integer greater than 1;
[0027] The processing unit is further used to obtain test feature data corresponding to the type of target feature data in the model test data, and input the test feature data into N trained classification prediction models respectively to obtain N groups of test result data;
[0028] The processing unit is further used to perform lift calculation operations on the N groups of test result data respectively to obtain test lift data corresponding to each group of test result data;
[0029] A determination unit is used to use the trained classification prediction model with the highest test lift data as the target classification prediction model.
[0030] In a possible design, in an implementation of another aspect of the embodiment of the present application, the acquisition unit may be specifically used to:
[0031] The consumption time interval, consumption frequency and consumption amount of the original variable data are obtained, and the consumption time interval, consumption frequency and consumption amount are used as derived characteristic data.
[0032] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0033] The business meaning derived data of the original variable data is obtained according to the business meaning, and the business meaning derived data is used as the derived feature data.
[0034] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0035] The original variable data is input into the integrated tree model for discretization and crossover, and the feature data obtained after discretization and crossover is used as the derived feature data.
[0036] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0037] Get the missing value ratio of derived feature data;
[0038] The derived feature data whose missing value ratio is greater than a preset missing value ratio threshold are eliminated, and the eliminated derived feature data are used as preprocessing feature data.
[0039] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0040] Get the single value ratio of the derived feature data;
[0041] The derived feature data whose single value ratio is greater than a preset single value ratio threshold are eliminated, and the eliminated derived feature data are used as preprocessing feature data.
[0042] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0043] Get missing values of derived feature data;
[0044] Fill the missing values according to the business meaning, and use the filled derived feature data as preprocessed feature data.
[0045] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0046] Obtain the statistical quantiles of the derived feature data;
[0047] The values of the statistical quantiles greater than the first quantile threshold are replaced by the values corresponding to the first quantile threshold, and the values of the statistical quantiles less than the second quantile threshold are replaced by the values corresponding to the second quantile threshold;
[0048] The replaced derived feature data is used as preprocessing feature data.
[0049] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0050] A data type conversion operation is performed on the derived feature data according to a preset data format, and the derived feature data after the data type conversion is used as preprocessing feature data.
[0051] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0052] Performing a first screening operation on the preprocessed feature data according to the information value to obtain a first screening feature;
[0053] Performing a second screening operation on the first screening feature according to the population stability index to obtain a second screening feature;
[0054] Performing a third screening operation on the second screening feature according to a stepwise regression method to obtain a third screening feature;
[0055] Performing a fourth screening operation on the third screening feature according to the sequence feature selection method to obtain a fourth screening feature;
[0056] A fifth screening operation is performed on the fourth screening feature according to the symbol consistency to obtain target feature data.
[0057] In a possible design, in an implementation of another aspect of the embodiment of the present application, the processing unit may be specifically used for:
[0058] Arrange N groups of test result data in descending order to obtain N groups of test sorting results;
[0059] Divide each group of test sorting results into equal parts according to the preset number of equal parts to obtain N groups of test binning results;
[0060] Calculate the bin lift data of N groups of test bin results respectively;
[0061] In the binned lift data corresponding to the same set of test result data, a preset number of target lift data are filtered according to the values from large to small;
[0062] The sum of the target lift data corresponding to each group of test result data is used as the test lift data corresponding to each group of test result data.
[0063] On the other hand, the present application provides an application device of a classification prediction model, comprising:
[0064] An acquisition unit, used to input the user information to be predicted into the classification prediction model, and output the prediction probability value through the classification prediction model;
[0065] A processing unit, configured to push product information to the predicted user if the predicted probability value is greater than a preset probability threshold;
[0066] The processing unit is also used to not execute the product information push operation if the predicted probability value is less than or equal to a preset probability threshold.
[0067] On the other hand, the present application provides a computer device, including: a memory, a processor, and a bus system;
[0068] Wherein, the memory is used to store programs;
[0069] The processor is used to implement the above-mentioned methods when executing the program in the memory;
[0070] The bus system is used to connect the memory and the processor so that the memory and the processor can communicate with each other.
[0071] Another aspect of the present application provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer is enabled to execute the above-mentioned methods.
[0072] It can be seen from the above technical solutions that the embodiments of the present application have the following beneficial effects:
[0073] The preprocessed feature data is obtained by performing feature derivation operation on the original variable data and preprocessing operation on the derived feature data; the target feature data is obtained by performing feature screening operation on the preprocessed feature data; the model training operation is performed on N original classification prediction models based on the target feature data and the label information corresponding to the target feature data; the test feature data corresponding to the type of the target feature data is obtained in the model test data, and the test feature data is input into N trained classification prediction models respectively, and the lift calculation operation is performed on N groups of test result data respectively to obtain the test lift data corresponding to each group of test result data; the trained classification prediction model with the highest test lift data is used as the target classification prediction model. Through the above method, the feature data suitable for classification prediction model training can be better obtained by performing feature derivation operation, feature preprocessing operation and feature screening operation on the original variable data, so as to better optimize the classification prediction model, and further, the test feature data corresponding to the type of the target feature data can be obtained to calculate the lift of the classification prediction model, so as to more accurately obtain the target classification prediction model that meets the business requirements, so that the information push object can be better predicted based on the target classification prediction model for information push, so as to improve the efficiency of information push. BRIEF DESCRIPTION OF THE DRAWINGS
[0074] Figure 1 This is a schematic diagram of the architecture of a data object control system in an embodiment of the present application;
[0075] Figure 2 It is a flow chart of an embodiment of the training method of the classification prediction model in the embodiment of the present application;
[0076] Figure 3 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0077] Figure 4 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0078] Figure 5is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0079] Figure 6 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0080] Figure 7 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0081] Figure 8 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0082] Fig. 9 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0083] Fig.10 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0084] Fig.11 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0085] Fig.12 is another embodiment flow chart of the training method of the classification prediction model in the embodiment of the present application;
[0086] Fig.13 This is a flow chart of an embodiment of the application method of the classification prediction model in the embodiment of the present application;
[0087] Fig.14 It is a schematic diagram of a principle flow chart of a training method for a classification prediction model in an embodiment of the present application;
[0088] Fig.15 It is a schematic diagram of the effect of a time window of the training method of the classification prediction model in the embodiment of the present application;
[0089] Fig.16 It is a schematic diagram of an embodiment of a training device for a classification prediction model in an embodiment of the present application;
[0090] Fig.17 This is a schematic diagram of an embodiment of an application device of a classification prediction model in an embodiment of the present application;
[0091] Fig.18 It is a schematic diagram of an embodiment of a computer device in an embodiment of the present application. DETAILED DESCRIPTION
[0092] The embodiments of the present application provide a training method, apparatus, and device storage medium for a classification prediction model, which are used to better obtain feature data suitable for classification prediction model training by performing feature derivation operations, feature preprocessing operations, and feature screening operations on original variable data, so as to better optimize the classification prediction model. Further, test feature data corresponding to the type of target feature data can be obtained to calculate the improvement of the classification prediction model, thereby more accurately obtaining a target classification prediction model that meets business requirements, so that the information push object can be better predicted based on the target classification prediction model for information push, so as to improve the efficiency of information push.
[0093] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "corresponding to" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0094] It is understandable that in the specific implementation of the present application, the original variable data, attribute information (such as the object virtual communication number, object nickname and other data mentioned in the subsequent embodiments) and other related data all need to be obtained through user authorization (i.e., user consent). In other words, when the above embodiments of the present application are applied to specific products or technologies, the methods and related functions provided by the embodiments of the present application are operated with the permission or consent of the user (the user can actively enable the functions provided by the embodiments of the present application), and the collection, use and processing of relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions.
[0095] It can be understood that the training method of the classification prediction model disclosed in the present application specifically relates to an Intelligent Vehicle Infrastructure Cooperative System (IVICS), and the Intelligent Vehicle Infrastructure Cooperative System is further introduced below. The Intelligent Vehicle Infrastructure Cooperative System is referred to as the Vehicle Infrastructure Cooperative System, which is a development direction of the Intelligent Transportation System (ITS). The vehicle-road cooperative system adopts advanced wireless communications and new generation Internet technologies to implement dynamic real-time information interaction between vehicles and roads in all directions, and carries out active vehicle safety control and road cooperative management based on the collection and integration of dynamic traffic information in all time and space, fully realizing the effective coordination of people, vehicles and roads, ensuring traffic safety, and improving traffic efficiency, thereby forming a safe, efficient and environmentally friendly road traffic system.
[0096] It is understandable that the training method of the classification prediction model disclosed in the present application also involves artificial intelligence (AI) technology, and artificial intelligence technology is further introduced below. Artificial intelligence is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that the machines have the functions of perception, reasoning and decision-making.
[0097] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0098] Secondly, natural language processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language people use in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.
[0099] Secondly, Machine Learning (ML) is a multi-disciplinary interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.
[0100] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. I believe that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.
[0101] It should be understood that the training method of the classification prediction model provided in this application can be applied to various scenarios, including but not limited to artificial intelligence, cloud technology, maps, smart transportation, etc., and is used to better obtain feature data that can be applied to the classification prediction model training by performing feature derivation operations, feature preprocessing operations, and feature screening operations on the original variable data, so as to better optimize the classification prediction model. Further, test feature data corresponding to the type of target feature data can be obtained to calculate the improvement of the classification prediction model, so as to more accurately obtain the target classification prediction model that meets the business requirements to realize information push. It can be applied to scenarios such as media information scenarios, social network platform information push scenarios, game information dissemination, etc.
[0102] In order to solve the above problems, this application proposes a training method for a classification prediction model. Figure 1 The data object control system shown is shown in Figure 1 , Figure 1 FIG. 1 is a schematic diagram of an architecture of a data object control system in an embodiment of the present application. Figure 1As shown, by performing feature derivation operation on the original variable data, and performing preprocessing operation on the derived feature data to obtain preprocessed feature data; performing feature screening operation on the preprocessed feature data to obtain target feature data; performing model training operation on N original classification prediction models based on the target feature data and the label information corresponding to the target feature data; obtaining test feature data corresponding to the type of the target feature data in the model test data, and inputting the test feature data into N trained classification prediction models respectively, performing lift calculation operation on N groups of test result data respectively, and obtaining test lift data corresponding to each group of test result data; taking the trained classification prediction model with the highest test lift data as the target classification prediction model. Through the above method, feature data applicable to classification prediction model training can be better obtained by performing feature derivation operation, feature preprocessing operation and feature screening operation on the original variable data, so as to better optimize the classification prediction model, and further, test feature data corresponding to the type of the target feature data can be obtained to calculate the lift of the classification prediction model, so as to more accurately obtain the target classification prediction model that meets the business requirements, so that the information push object can be better predicted based on the target classification prediction model for information push, so as to improve the efficiency of information push.
[0103] Understandably, Figure 1 The terminal devices are shown in the figure. In actual scenarios, more types of terminal devices can participate in the data processing process. Terminal devices also include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, car terminals, etc. The specific number and types depend on the actual scenario and are not limited here. In addition, Figure 1 One server is shown in the figure, but in actual scenarios, multiple servers may also be involved, especially in the scenario of multi-model training interaction. The number of servers depends on the actual scenario and is not limited here.
[0104] It should be noted that in this embodiment, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. The terminal device and the server can be directly or indirectly connected through wired or wireless communication, and the terminal device and the server can be connected to form a blockchain network, which is not limited in this application.
[0105] Combined with the above introduction, the following will introduce the training method of the classification prediction model in this application, please refer to Figure 2 , an embodiment of a training method for a classification prediction model in an embodiment of the present application includes:
[0106] In step S101, original sample data carrying label information is obtained, wherein the original sample data includes model training data and model testing data;
[0107] It is understandable that in a virtual transaction business scenario, the multiple business products launched are mainly divided into two types: revolving business products and installment business products. Revolving business products are products that can extract asset data multiple times after a product application is approved, and installment business products are products that apply for a single asset data for a single product. The two business products have large differences in product positioning, risk, and asset thickness. In order to increase the activity of product purchase objects and the stickiness of the business products, it is necessary to cross-market the existing stock product purchase objects in the current scenario, such as guiding product purchase objects that only apply for revolving business products to apply for installment business products, or guiding product purchase objects that only apply for installment business products to apply for revolving business products. Since the asset thickness of installment business products is usually lower than that of revolving business products, this embodiment guides product purchase objects that only apply for installment business products to apply for revolving business products. Therefore, it is necessary to accurately find product purchase objects that are potentially willing to apply for revolving business products after applying for installment business products.
[0108] Therefore, this embodiment can obtain the original sample data carrying label information to filter out feature data that better performs model training and optimization, so as to obtain the target classification prediction model, so that the information push object can be better predicted based on the target classification prediction model for information push, so as to improve the efficiency of information push.
[0109] Among them, the original sample data includes model training data and model testing data.
[0110] Specifically, the original sample data carrying label information can be obtained by combining actual business and considering several important time points, such as Fig.15 The observation point shown: refers to the critical point that divides the observation period and the performance period. The period before the observation point becomes the observation period, and the period after the observation point becomes the performance period.
[0111] For example, the observation period and performance period can be defined as, e.g. Fig.15As shown in the figure, assuming that a product purchase object applies for an installment business product on December 31, 2022, the time window from the launch time of the installment business product (such as 20221001) to the time when the product purchase object applies for the installment business product (20221001-20221231) is defined as the observation period, and the future period after applying for the installment business product, such as one month (20230101-20230131), is defined as the performance period. The reason why the time window is defined as the next month is that the variable X of the product purchase object usually does not change within a month.
[0112] Furthermore, regarding the definition of label information: it refers to the product purchase object who has not applied for the revolving business product during the observation period, and has applied for the revolving business product within one month after applying for the installment business product (within the performance period) is the target product purchase object, that is, the positive sample, recorded as label information Y=1; similarly, the product purchase object who has not applied for the revolving business product during the observation period, and has not applied for the revolving business product within one month after applying for the installment business product (within the performance period) is the non-target product purchase object, that is, the negative sample, recorded as label information Y=0.
[0113] Furthermore, for example, the process of selecting 4532 original sample data and label information from 8331 samples is shown in Table 1 below:
[0114] Table 1
[0115]
[0116]
[0117] Furthermore, after obtaining the original sample data and the label information corresponding to each original sample, in order to better evaluate the discrimination, stability, robustness, etc. of the model, the present embodiment can randomly divide the screened original sample data into a training set and a test set at a ratio of 6:4, i.e., model training data and model test data. Among them, model building and parameter adjustment are performed on the model training data, while the model test data does not participate in the modeling process and is only used to evaluate the prediction performance of the model.
[0118] For example, the model training data and model test data are divided from 4532 original sample data as shown in Table 2 below:
[0119]
[0120] In step S102, original variable data corresponding to the model training data is obtained;
[0121] In an embodiment of the present application, after acquiring the model training data, the original variable data corresponding to the model training data can be acquired, so that feature data suitable for classification prediction model training can be better acquired by subsequently performing feature derivation operations, feature preprocessing operations, and feature screening operations on the original variable data, so as to better optimize the classification prediction model.
[0122] The original variable data refers to the attribute information and behavior information generated by the product purchase object when applying for the business product. For example, it can be understood that the variables corresponding to the attribute information and behavior information include the following categories:
[0123] ① Basic attribute information of the product purchase object: such as nickname, virtual communication number, product type preference (such as installment business products or recurring business products, etc.), product business cycle preference (such as once a week, once a month, etc.), etc.;
[0124] ②Transaction data of product purchase objects in business scenarios: such as pre-stored asset values, purchased asset values, asset data transaction details, etc.;
[0125] ③ Behavioral data of product purchasers in other business scenarios: such as the time and frequency of product purchasers’ application for business products in other business institutions;
[0126] ④ Special attribute information of the product purchase object: such as the region where the product purchase object purchases the business product, etc.
[0127] Specifically, Fig.14 As shown, after obtaining the model training data, the original variable data corresponding to the model training data can be obtained, that is, suitable variables can be found for modeling. Therefore, the attribute information and behavior information associated with the sample identifier (for example, user ID) in the model training data can be screened from the database or big data platform to obtain the original variable data.
[0128] In step S103, a feature derivation operation is performed on the original variable data to obtain derived feature data;
[0129] In an embodiment of the present application, after obtaining the original variable data, in order to better obtain feature data suitable for classification prediction model training and to better optimize the classification prediction model, this embodiment can perform sample feature expansion by performing feature derivation operations on the original variable data to mine more types or features of derived feature data.
[0130] Specifically, Fig.14 As shown, after obtaining the original variable data, the feature derivation operation (such as Fig.14The feature derivation indicated in the figure) can specifically be to expand the sample features by using one or more of the algorithms such as the data analysis method, i.e., the RFM method, the business meaning derivation algorithm, the ensemble tree model feature derivation algorithm, etc., so as to mine more types or characteristics of derived feature data.
[0131] In step S104, a preprocessing operation is performed on the derived feature data to obtain preprocessed feature data;
[0132] In an embodiment of the present application, after the derived feature data is expanded, in order to further maintain the validity of the feature and facilitate computer recognition and processing, the present embodiment may perform a preprocessing operation on the derived feature data to obtain preprocessed feature data with a standardized and unified format for easy recognition.
[0133] Specifically, after the derived feature data is expanded, the derived feature data can be preprocessed (such as Fig.14 Specifically, one or more of the missing value ratio processing algorithm, single value ratio processing algorithm, missing value processing algorithm, extreme value processing, and data type conversion algorithms shown in Table 3 below may be used to perform feature preprocessing operations to obtain preprocessed feature data with a standardized and unified format for easy identification:
[0134] Table 3
[0135] Features Feature preprocessing Missing value ratio processing If the missing value ratio is > 90%, remove Single value ratio processing Single value ratio> 90%, removed Missing Value Handling Missing values are filled based on business meaning Extreme value processing Replace those greater than P99 with P99, and those less than P1 with P1 Data type conversion Unified data format, etc.
[0136] In step S105, a feature screening operation is performed on the preprocessed feature data to obtain target feature data;
[0137] In the embodiment of the present application, since the performance of each variable on the model training data is analyzed, it can be understood as the ability of the variable to distinguish between positive samples and negative samples in the model training data. It can be understood that the greater the distinction between positive samples and negative samples by the variable, the more obvious the improvement of the effect of the variable on the model. Therefore, after obtaining the preprocessed feature data, the preprocessed feature data can be further subjected to feature screening operations to obtain target feature data that has a more obvious improvement on the effect of the model.
[0138] Specifically, Fig.14 As shown, after obtaining the preprocessing feature data, the preprocessing feature data can be further subjected to feature screening operations (such as Fig.14Specifically, the first screening operation can be performed on the preprocessed feature data according to the information value (for example, the information value or the amount of information, that is, the size of the IV value, to perform feature screening) to obtain the first screening feature; and then the second screening operation can be performed on the first screening feature according to the population stability index (for example, the population stability index index, that is, the proportion of PSI index for feature screening) to obtain the second screening feature; according to the stepwise regression method, the second screening feature is subjected to a third screening operation (for example, any one of the Filter-forward stepwise regression feature screening algorithm and the Filter-bidirectional stepwise regression feature screening algorithm is used to obtain the test statistic p value of each feature for feature screening, or two stepwise regression algorithms are simultaneously executed to obtain the test statistic p value of each feature and then the union is taken for feature screening) to obtain the third screening feature; further, according to the sequence feature selection method, the third screening feature is subjected to a fourth screening operation (for example, any one of the Wrapper-sequence forward selection algorithm and the Wrapper-sequence floating forward selection algorithm is used for feature screening, or two stepwise regression algorithms are simultaneously executed and then the union is taken for feature screening) to obtain the fourth screening feature; then, according to the sign consistency, the fourth screening feature is subjected to a fifth screening operation (for example, the sign consistency of numerical values such as the test statistic p value or the VIF value is considered) to obtain the target feature data, that is, the target variable X.
[0139] In step S106, N original classification prediction models are constructed according to the model type, and model training operations are performed on the original classification prediction models based on the target feature data and the label information corresponding to the target feature data, to obtain N trained classification prediction models, where N is an integer greater than 1;
[0140] In an embodiment of the present application, N original classification prediction models can be constructed according to the business type or model type, and then the target feature data and the label information corresponding to the target feature data can be used to perform model training operations on the original classification prediction models to obtain N trained classification prediction models, so that the effects of the N trained classification prediction models can be evaluated later to further obtain the target classification prediction model that is compatible with the target business.
[0141] Specifically, before acquiring the target feature data, this embodiment can construct N original classification prediction models according to the business type or model type, such as logistic regression, decision tree, extreme gradient boosting tree XGBoost and regression tree GBDT model, etc., without specific limitation here.
[0142] Furthermore, after obtaining the target feature data, the target feature data (i.e., target variable X, for example, object nickname or object virtual communication number, etc.) can be used as input data and label information (sample label Y value) as result data to perform model training operations on each original classification prediction model respectively. Specifically, for each original classification prediction model, the target feature data (i.e., target variable X, for example, object nickname or object virtual communication number, etc.) can be used as model input, and the probability value corresponding to each feature X (i.e., each sample) can be output by the original classification prediction model. Then, according to the loss function corresponding to each original classification prediction model, for example, the loss function corresponding to the logistic regression model is cross entropy, that is, the loss value can be calculated based on the probability value corresponding to each feature X (i.e., each sample) and the label information, and the model parameters of the original classification prediction model can be updated based on the loss value. Similarly, N trained classification prediction models can be obtained.
[0143] In step S107, test feature data corresponding to the type of target feature data is obtained from the model test data, and the test feature data is respectively input into N trained classification prediction models to obtain N groups of test result data;
[0144] Specifically, after obtaining the target feature data, test feature data corresponding to the type of the target feature data (for example, the object nickname or the object virtual communication number, etc.) can be obtained in the model test data, and then the test feature data can be input into N trained classification prediction models respectively, so as to output N groups of test result data (i.e., prediction probability values) through the N trained classification prediction models respectively, so that the test improvement data corresponding to each trained classification prediction model can be calculated based on the N groups of test result data, so as to better evaluate the effect of each trained classification prediction model.
[0145] In step S108, lift calculation operations are performed on the N groups of test result data respectively to obtain test lift data corresponding to each group of test result data;
[0146] In an embodiment of the present application, after obtaining N groups of test result data, in order to be able to more intuitively and quickly reflect the effect evaluation of each trained classification prediction model, this embodiment can perform lift calculation operations on the N groups of test result data respectively to obtain the test lift data corresponding to each group of test result data, and use the test lift data to express the effect evaluation results of each trained classification prediction model.
[0147] Among them, the lift calculation operation usually adopts the lift algorithm, where lift refers to a metric for evaluating whether a prediction model is effective; it measures how much better a model (or rule) is at predicting the "response" in the target than random selection, usually with 1 as the boundary. It can be understood that a lift greater than 1 means that the model or rule captures more "responses" than random selection, a lift equal to 1 means that the model's performance is independent of random selection, and a lift less than 1 means that the model or rule captures less "responses" than random selection.
[0148] Specifically, after obtaining N groups of test result data, the output probability value of each sample in each group of test result data is sorted in descending order, and then the modeling sample is divided into x equal parts (for example, 8 equal parts, not specifically limited here), and the ratio of the proportion of positive samples in each group of test result data and the proportion of positive samples in all samples in the N groups of test result data in each equal part of the sample is checked respectively. This ratio is the test lift data corresponding to the group of test result data, that is, the lift.
[0149] It is understandable that in business scenarios, we do not pay attention to all x-divided data, but only the top 1-3 divided data, which have the highest response rate. The following table 4 shows the effect of the regression tree GBDT model on the model training data, and the following table 5 shows the effect of the regression tree GBDT model on the 8-divided bins on the model test data:
[0150] Table 4
[0151] Box serial number Number of negative samples Number of positive samples Total number of samples Positive sample ratio Lift value 1 173 236 409 58% 3.41 2 293 120 413 29% 1.72 3 347 65 412 16% 0.93 4 347 63 410 15% 0.91 5 365 34 399 9% 0.50 6 386 22 408 5% 0.32 7 329 11 340 3% 0.19 8 494 6 500 1% 0.07 Total 2734 557 3291 17% 1.00
[0152] Table 5
[0153]
[0154]
[0155] It can be seen from Tables 4 and 5 above that not only is the lift value of the top 1 bin in the model training data higher than 2, but the lift value of the top 1 bin in the model test data is also higher than 2, which can be used to indicate that the model effect is good and the model is relatively stable. Therefore, the lift value of each trained classification prediction model in the top 1 bin can be used as the test lift data corresponding to the model.
[0156] In step S109, the trained classification prediction model with the highest test lift data is used as the target classification prediction model.
[0157] In an embodiment of the present application, after obtaining the test lift data corresponding to each group of test result data, the test lift data corresponding to each group of test result data can be compared pairwise to obtain the test lift data with the largest value, and then the trained classification prediction model with the highest test lift data can be used as the target classification prediction model.
[0158] In an embodiment of the present application, a training method for a classification prediction model is provided. Through the above-mentioned method, feature data suitable for classification prediction model training can be better obtained by performing feature derivation operations, feature preprocessing operations and feature screening operations on the original variable data, so as to better optimize the classification prediction model. Further, test feature data corresponding to the type of target feature data can be obtained to calculate the improvement degree of the classification prediction model, so as to more accurately obtain the target classification prediction model that meets the business requirements, so that the information push object can be better predicted based on the target classification prediction model for information push, so as to improve the efficiency of information push.
[0159] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Figure 3 As shown, in step S103, feature derivation operation is performed on the original variable data to obtain derived feature data, including:
[0160] In step S301, the consumption time interval, consumption frequency and consumption amount of the original variable data are obtained, and the consumption time interval, consumption frequency and consumption amount are used as derived feature data.
[0161] In an embodiment of the present application, after obtaining the original variable data, in order to better obtain feature data suitable for classification prediction model training to better optimize the classification prediction model, this embodiment can perform feature derivation operations on the original variable data, that is, by obtaining the consumption time interval, consumption frequency and consumption amount of the original variable data, and using the consumption time interval, consumption frequency and consumption amount as derived feature data of more types or features mined.
[0162] Specifically, Fig.14 As shown, after obtaining the original variable data, the feature derivation operation (such as Fig.14The feature derivation indicated as shown in the figure) can be specifically carried out by adopting, for example, a data analysis method, namely, the RFM method, wherein RFM is used to represent the most recent consumption time (R), consumption frequency (F), and consumption amount (M), respectively. For example, the behavior information of the product purchase object in applying for business products in other business institutions is derived into the number of days since the most recent application for business products (R), namely, the consumption time interval, the number of times the business products are applied in the past three months (F), namely, the consumption frequency, and the amount of the business products applied in the past three months (M), namely, the consumption amount, as derived feature data.
[0163] Furthermore, based on the derived feature data obtained by the data analysis method, namely the RFM method, one or more algorithms such as the business meaning derivation algorithm and the ensemble tree model feature derivation algorithm can be selectively combined to expand sample features in order to mine more types or characteristics of derived feature data.
[0164] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Figure 4 As shown, in step S103, feature derivation operation is performed on the original variable data to obtain derived feature data, including:
[0165] In step S401, business meaning derived data of original variable data is obtained according to the business meaning, and the business meaning derived data is used as derived feature data.
[0166] In an embodiment of the present application, after obtaining the original variable data, in order to better obtain feature data suitable for classification prediction model training and to better optimize the classification prediction model, the present embodiment can perform feature derivation operations on the original variable data, that is, obtain business meaning derived data of the original variable data according to the business meaning, and use the business meaning derived data as derived feature data of more types or features mined.
[0167] Specifically, Fig.14 As shown, after obtaining the original variable data, the feature derivation operation (such as Fig.14 The feature derivation indicated in the figure may specifically be, for example, a derivation algorithm based on business meaning, that is, deriving variables according to business meaning (for example, basic attribute information of the product purchase object: such as nickname, virtual communication number, product type preference, product business cycle preference, etc.) and business management personnel's business experience, such as deriving whether the product purchase object likes to buy high-quality products, the proportion of the product purchase object's liabilities in installment business products of other institutions, the proportion of the product purchase object's liabilities in revolving business products of other institutions, etc., as derived feature data.
[0168] Furthermore, based on the derived feature data obtained based on the business meaning derivation algorithm, the sample features can be expanded by selectively combining one or more algorithms such as the data analysis method, namely the RFM method, the integrated tree model feature derivation algorithm, etc., so as to mine more types or characteristics of derived feature data.
[0169] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Figure 5 As shown, step S103 performs feature derivation operation on the original variable data to obtain derived feature data, including:
[0170] In step S501, the original variable data is input into the integrated tree model for discretization and crossover, and the feature data obtained after discretization and crossover is used as derived feature data.
[0171] In an embodiment of the present application, after obtaining the original variable data, in order to better obtain feature data suitable for classification prediction model training and to better optimize the classification prediction model, this embodiment can perform feature derivation operations on the original variable data, that is, input the original variable data into the integrated tree model for discretization and crossover, and use the feature data obtained after discretization and crossover as derived feature data of more types or features mined.
[0172] Specifically, Fig.14 As shown, after obtaining the original variable data, the feature derivation operation (such as Fig.14 The feature derivation indicated in the figure) can be specifically implemented by, for example, adopting a feature derivation algorithm based on an integrated tree model, that is, inputting the original variable data into the integrated tree model for feature discretization and crossover, and automatically constructing the feature data obtained after discretization and crossover, and the constructed feature data is the derived feature data.
[0173] Furthermore, based on the derived feature data obtained based on the integrated tree model feature derivation algorithm, the sample features can be expanded by selectively combining one or more algorithms such as the data analysis method, namely the RFM method, the business meaning derivation algorithm, etc., so as to mine more types or characteristics of derived feature data.
[0174] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Figure 6 As shown, step S104 performs a preprocessing operation on the derived features to obtain preprocessed feature data, including:
[0175] In step S601, the missing value ratio of the derived feature data is obtained;
[0176] In step S602, the derived feature data with a missing value ratio greater than a preset missing value ratio threshold is eliminated, and the eliminated derived feature data is used as preprocessed feature data.
[0177] In an embodiment of the present application, after the derived feature data is expanded, in order to further maintain the validity of the feature and facilitate computer recognition and processing, the present embodiment can perform a preprocessing operation on the derived feature data, that is, by obtaining the missing value ratio of the derived feature data, and eliminating the derived feature data whose missing value ratio is less than a preset missing value ratio threshold, and then use the eliminated derived feature data as preprocessed feature data with higher numerical validity.
[0178] Specifically, after the derived feature data is expanded, the derived feature data can be preprocessed (such as Fig.14 The feature preprocessing indicated in the figure) may be, for example, the missing value ratio processing algorithm indicated in Table 3 above. Specifically, the derived feature data, such as the variable "nickname", if the values corresponding to all product purchase objects are null values, then the nickname variable is missing values, and the missing value ratio is 100%. The missing value ratio of the derived feature data is compared with the preset missing value ratio threshold (such as 90%). It can be seen that if the missing value ratio is 100% greater than 90%, the nickname variable has no effect on distinguishing product purchase objects, and the nickname variable can be eliminated. Furthermore, the derived feature data after elimination can be used as preprocessed feature data.
[0179] Furthermore, on the basis of obtaining preprocessed feature data based on the missing value ratio processing algorithm, one or more algorithms such as the single value ratio processing algorithm, the missing value processing algorithm, the extreme value processing algorithm, and the data type conversion algorithm can be selectively combined to perform feature preprocessing operations to obtain preprocessed feature data with a standardized and unified format for easy identification.
[0180] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Figure 7 As shown, step S104 performs a preprocessing operation on the derived features to obtain preprocessed feature data, including:
[0181] In step S701, a single value ratio of the derived feature data is obtained;
[0182] In step S702, the derived feature data whose single value ratio is greater than a preset single value ratio threshold is eliminated, and the eliminated derived feature data is used as pre-processed feature data.
[0183] In an embodiment of the present application, after the derived feature data is expanded, in order to further maintain the validity of the feature and facilitate computer recognition and processing, the present embodiment can perform a preprocessing operation on the derived feature data, that is, by obtaining the single value ratio of the derived feature data, and eliminating the derived feature data whose single value ratio is greater than a preset single value ratio threshold, and then use the eliminated derived feature data as preprocessed feature data with higher numerical validity.
[0184] Specifically, after the derived feature data is expanded, the derived feature data can be preprocessed (such as Fig.14 The feature preprocessing indicated in the figure may be, for example, the single value ratio processing algorithm indicated in Table 3 above. Specifically, the derived feature data may include, for example, the variable "virtual communication number". If the value of all product purchase objects is 123456 (that is, the virtual communication numbers of all product purchase objects are 123456), then the variable virtual communication number has a single value of 123456, and the single value ratio is 100%. The single value ratio of the derived feature data is compared with a preset single value ratio threshold (such as 90%). It can be seen that if the single value ratio is 100% greater than 90%, the variable virtual communication number has no effect on distinguishing product purchase objects. The variable virtual communication number can be eliminated, and the derived feature data after elimination can be used as preprocessing feature data.
[0185] Furthermore, on the basis of obtaining preprocessed feature data based on a single value ratio processing algorithm, one or more of the missing value ratio processing algorithm, missing value processing algorithm, extreme value processing algorithm, and data type conversion algorithms can be selectively combined to perform feature preprocessing operations to obtain preprocessed feature data with a standardized and unified format for easy identification.
[0186] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Figure 8 As shown, step S104 performs a preprocessing operation on the derived features to obtain preprocessed feature data, including:
[0187] In step S801, missing values of derived feature data are obtained;
[0188] In step S802, the missing values are filled in according to the business meaning, and the derived feature data after filling is used as preprocessing feature data.
[0189] In an embodiment of the present application, after the derived feature data is expanded, in order to further maintain the validity of the feature and facilitate computer recognition and processing, the present embodiment can perform a preprocessing operation on the derived feature data, that is, by obtaining the missing values of the derived feature data, filling the missing values according to the business meaning, and then using the filled derived feature data as preprocessed feature data with higher numerical validity.
[0190] Specifically, after the derived feature data is expanded, the derived feature data can be preprocessed (such as Fig.14 The feature preprocessing indicated in the figure may be, for example, the missing value processing algorithm indicated in Table 3 above. Specifically, the derived feature data may be a variable such as "product business cycle preference". If the values corresponding to most product purchase objects are null values, then the variable of product business cycle preference is a missing value. The missing values are filled in according to the business meaning (for example, the basic attribute information of the product purchase object, such as the cycle is purchased once a month). Then, the filled derived feature data is used as the preprocessed feature data with higher numerical validity.
[0191] Furthermore, on the basis of obtaining preprocessed feature data based on the missing value processing algorithm, one or more algorithms such as single value ratio processing algorithm, missing value ratio processing algorithm, extreme value processing algorithm and data type conversion algorithm can be selectively combined to perform feature preprocessing operations to obtain preprocessed feature data with a standardized and unified format for easy identification.
[0192] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Fig. 9 As shown, step S104 performs a preprocessing operation on the derived features to obtain preprocessed feature data, including:
[0193] In step S901, the statistical quantile of the derived feature data is obtained;
[0194] In step S902, the values of the statistical quantiles greater than the first quantile threshold are replaced by the values corresponding to the first quantile threshold, and the values of the statistical quantiles less than the second quantile threshold are replaced by the values corresponding to the second quantile threshold;
[0195] In step S903, the derived feature data after replacement is used as pre-processed feature data.
[0196] In an embodiment of the present application, after the derived feature data is expanded, in order to further maintain the validity of the feature and facilitate computer recognition and processing, the present embodiment can perform a preprocessing operation on the derived feature data, that is, by obtaining the statistical quantiles of the derived feature data, and then, replacing the values of the statistical quantiles greater than the first quantile threshold with the values corresponding to the first quantile threshold, and replacing the values of the statistical quantiles less than the second quantile threshold with the values corresponding to the second quantile threshold, and then using the replaced derived feature data as preprocessed feature data with higher numerical validity.
[0197] Specifically, after the derived feature data is expanded, the derived feature data can be preprocessed (such as Fig.14 The feature preprocessing indicated in the figure may be, for example, the extreme value processing algorithm indicated in Table 3 above, which may be specifically to obtain the statistical quantile of the derived feature data, and then respectively compare the statistical quantile of the derived feature data with the statistical quantile of the first quantile threshold (such as P99) and the statistical quantile of the second quantile threshold (such as P1), then the values of the statistical quantiles greater than the first quantile threshold may be replaced by the values corresponding to the first quantile threshold (that is, those greater than P99 are replaced with P99), and the values of the statistical quantiles less than the second quantile threshold may be replaced by the values corresponding to the second quantile threshold (that is, those less than P1 are replaced with P1), and then the replaced derived feature data are used as preprocessed feature data with higher numerical validity.
[0198] Among them, P1 and P99 are statistical quantiles, that is, after sorting the samples in ascending order according to their numerical values, the samples are divided into 100 equal parts. Then P1 represents the value corresponding to 1%, P99 represents the value corresponding to 99%, and P50 is the median.
[0199] Furthermore, on the basis of obtaining pre-processed feature data based on the extreme value processing algorithm, one or more algorithms such as the missing value ratio processing algorithm, the single value ratio processing algorithm, the missing value processing algorithm, and the data type conversion algorithm can be selectively combined to perform feature preprocessing operations to obtain pre-processed feature data with a standardized and unified format for easy identification.
[0200] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Fig.10 As shown, step S104 performs a preprocessing operation on the derived features to obtain preprocessed feature data, including:
[0201] In step S1001, a data type conversion operation is performed on the derived feature data according to a preset data format, and the derived feature data after the data type conversion is used as preprocessing feature data.
[0202] In an embodiment of the present application, after the derived feature data is expanded, in order to further maintain the validity of the feature and facilitate computer recognition and processing, the present embodiment can perform a preprocessing operation on the derived feature data, that is, by performing a data type conversion operation on the derived feature data according to a preset data format, and then using the derived feature data after the data type conversion as preprocessed feature data with a unified format specification.
[0203] Specifically, after the derived feature data is expanded, the derived feature data can be preprocessed (such as Fig.14 The feature preprocessing shown in the figure can be specifically adopted, for example, the data type conversion algorithm shown in Table 3 above, and specifically the data type conversion operation can be performed on the derived feature data according to the preset data format, wherein. For the classification prediction model, it is generally necessary to convert the data format into a numerical variable to facilitate the model calculation. For example, for the variable of city, the value is city A, city B, etc. This kind of variable nominal variable cannot be directly input into the classification prediction model for calculation. Therefore, when this embodiment traverses to this kind of variable, it is usually necessary to convert it into a numerical variable. For example, "Is it city A?", after being converted into a numerical variable, the corresponding value is "yes = 1" and "no = 0"; or, "Is it city B?", after being converted into a numerical variable, the corresponding value is "yes = 1" and "no = 0", etc.
[0204] Furthermore, on the basis of obtaining preprocessed feature data based on the missing value ratio processing algorithm, one or more of the missing value ratio processing algorithm, the single value ratio processing algorithm, the missing value processing algorithm and the extreme value processing algorithm can be selectively combined to perform feature preprocessing operations to obtain preprocessed feature data with a standardized and unified format for easy identification.
[0205] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Fig.11 As shown, step S105 performs a feature screening operation on the pre-processed feature data to obtain target feature data, including:
[0206] In step S1101, a first screening operation is performed on the preprocessed feature data according to the information value to obtain a first screening feature;
[0207] In step S1102, a second screening operation is performed on the first screening feature according to the population stability index to obtain a second screening feature;
[0208] In step S1103, a third screening operation is performed on the second screening feature according to the stepwise regression method to obtain a third screening feature;
[0209] In step S1104, a fourth screening operation is performed on the third screening feature according to the sequence feature selection method to obtain a fourth screening feature;
[0210] In step S1105, a fifth screening operation is performed on the fourth screening feature according to the symbol consistency to obtain target feature data.
[0211] In the embodiment of the present application, since the performance of each variable on the model training data is analyzed, it can be understood as the ability of the variable to distinguish between positive samples and negative samples in the model training data. It can be understood that the greater the distinction between positive samples and negative samples by the variable, the more obvious the effect of the variable on the model is improved. Therefore, after obtaining the preprocessed feature data, the preprocessed feature data can be further subjected to feature screening operations, that is, a first screening operation can be performed on the preprocessed feature data according to the information value to obtain the first screening feature, and then a second screening operation can be performed on the first screening feature according to the population stability index to obtain the second screening feature, and a third screening operation can be performed on the second screening feature according to the stepwise regression method to obtain the third screening feature, and then a fourth screening operation can be performed on the third screening feature according to the sequence feature selection method to obtain the fourth screening feature, and then a fifth screening operation can be performed on the fourth screening feature according to the symbol consistency to obtain target feature data with more obvious improvement on the effect of the model.
[0212] Among them, as shown in Table 6 below, the first screening operation corresponds to step 1 of using IV value (i.e., information value or information quantity Information Value, IV value is used to measure the predictive ability of independent variables on dependent variables) for feature screening, that is, the size of the IV value is compared with the preset comparison threshold, usually screening features with IV values greater than or equal to 0.01. As the first screening feature, it can quickly remove a large amount of feature data that is useless to the model, or feature data with very low correlation with label information Y, and can quickly and effectively reduce the number of features.
[0213] Further, as shown in Table 6 below, the second screening operation corresponds to the use of the PSI index (i.e., population stability index, PSI index, which is usually used to measure the stability of a model or feature) in step 2 to screen for feature stability, that is, the proportion of the PSI index is compared with a preset comparison threshold, and usually the features with a PSI index proportion of less than 10% are screened as the second screening features.
[0214] Further, since both the IV value and the PSI index only consider the relationship between a single feature X and the tag information Y, and do not consider the correlation between the features, feature screening cannot be overly dependent on the IV value or the PSI index, so this embodiment further uses the method of stepwise regression and sequence feature selection to further screen the features. Wherein, as shown in Table 6 below, the third screening operation corresponds to the use of stepwise regression (i.e., Filter-forward stepwise regression feature screening) in step 3-1, by considering the test statistic p value of each feature after multiple feature modeling, which belongs to the filtering feature screening method, the test statistic p value is compared with the preset statistical threshold, and the features with a P value less than 0.05 are selected, and / or the use of stepwise regression (i.e., Filter-bidirectional stepwise regression feature screening algorithm) in step 3-2, the test statistic p value is compared with the preset statistical threshold, the features with a P value less than 0.05 are selected, and the features with a P value greater than 0.05 are eliminated, so as to summarize and obtain the third screening feature.
[0215] Further, as shown in Table 6 below, the fourth screening operation corresponds to step 4-1 using sequence feature screening (i.e., Wrapper-sequence forward selection) by considering the effect of the entire model after multiple feature modeling, which is a wrapper-type feature screening method to select 30 optimal feature combinations, and / or step 4-2 using sequence feature screening (i.e., Wrapper-sequence floating forward selection) to select 3-50 optimal feature combinations.
[0216] Further, as shown in Table 6 below, the fifth screening operation corresponds to the use of sign consistency in step 5 (considering the sign consistency of numerical values such as the test statistic p value or the VIF value), that is, eliminating features with a P value greater than 0.01 and eliminating features with a VIF value greater than 2 to obtain the target feature data, that is, the target variable X.
[0217] Table 6
[0218]
[0219] Specifically, Fig.14 As shown, after obtaining the preprocessing feature data, the preprocessing feature data can be further subjected to feature screening operations (such as Fig.14Specifically, the first screening operation can be performed on the preprocessed feature data according to the information value (for example, the information value or the amount of information, that is, the size of the IV value, to perform feature screening) to obtain the first screening feature; and then the second screening operation can be performed on the first screening feature according to the population stability index (for example, the population stability index index, that is, the proportion of PSI index for feature screening) to obtain the second screening feature; according to the stepwise regression method, the second screening feature is subjected to a third screening operation (for example, any one of the Filter-forward stepwise regression feature screening algorithm and the Filter-bidirectional stepwise regression feature screening algorithm is used to obtain the test statistic p value of each feature for feature screening, or two stepwise regression algorithms are simultaneously executed to obtain the test statistic p value of each feature and then the union is taken for feature screening) to obtain the third screening feature; further, according to the sequence feature selection method, the third screening feature is subjected to a fourth screening operation (for example, any one of the Wrapper-sequence forward selection algorithm and the Wrapper-sequence floating forward selection algorithm is used for feature screening, or two stepwise regression algorithms are simultaneously executed and then the union is taken for feature screening) to obtain the fourth screening feature; then, according to the sign consistency, the fourth screening feature is subjected to a fifth screening operation (for example, the sign consistency of numerical values such as the test statistic p value or the VIF value is considered) to obtain the target feature data, that is, the target variable X.
[0220] Optionally, in the above Figure 2 Based on the corresponding embodiment, in another optional embodiment of the training method of the classification prediction model provided in the embodiment of the present application, as Fig.12 As shown, step S108 performs lift calculation operations on N groups of test result data respectively to obtain test lift data corresponding to each group of test result data, including:
[0221] In step S1201, N groups of test result data are sorted in descending order to obtain N groups of test sorting results;
[0222] In step S1202, each group of test sorting results is equally divided according to a preset number of equal parts to obtain N groups of test binning results;
[0223] In step S1203, the binning lift data of the N groups of test binning results are calculated respectively;
[0224] In step S1204, in the binned lift data corresponding to the same set of test result data, a preset number of target lift data are filtered according to the values from large to small;
[0225] In step S1205, the sum of the target lift data corresponding to each group of test result data is used as the test lift data corresponding to each group of test result data.
[0226] In an embodiment of the present application, after obtaining N groups of test result data, in order to be able to more intuitively and quickly reflect the effect evaluation of each trained classification prediction model, this embodiment can perform lift calculation operations on the N groups of test result data respectively, that is, by performing descending sorting operations on the N groups of test result data respectively to obtain N groups of test sorting results, and dividing each group of test sorting results into equal parts according to a preset number of equal parts to obtain N groups of test bin results, and then calculating the bin lift data of the N groups of test bin results respectively, and then, in the bin lift data corresponding to the same group of test result data, a preset number of target lift data are filtered from large to small according to the numerical value, and the sum of the target lift data corresponding to each group of test result data is used as the test lift data corresponding to each group of test result data, and the test lift data is used to express the effect evaluation results of each trained classification prediction model.
[0227] Specifically, after obtaining N groups of test result data, the N groups of test result data are respectively arranged in descending order to obtain N groups of test sorting results. Specifically, the output probability value of each sample in each group of test result data is sorted in descending order, and the obtained sorting result is used as the test sorting result corresponding to each group of test result data to obtain N groups of test sorting results.
[0228] Furthermore, each group of test sorting results is divided equally according to a preset number of equal parts to obtain N groups of test binning results. Specifically, each group of test sorting results can be divided into x equal parts (for example, 8 equal parts, not specifically limited here) to obtain the test binning results corresponding to each group of test sorting results, so as to obtain N groups of test binning results.
[0229] Furthermore, the bin lift data of the N groups of test bin results are calculated respectively, that is, for each group of test bin results in the N groups of test bin results, such as a group of test bin results shown in Table 5 above, the ratio of the proportion of positive samples in the group of test bin results to the proportion of positive samples in all samples in the N groups of test bin results is calculated, and the ratio is obtained as the bin lift data corresponding to the group of test bin results, that is, the lift values corresponding to bin numbers 1 to 8, respectively.
[0230] Further, among the bin lift data corresponding to the same group of test result data, a preset number of target lift data are filtered according to the values from large to small. For example, for the bin lift data corresponding to the same group of test result data as shown in Table 5 above, that is, the lift values corresponding to bin number 1 to bin number 8, respectively, a preset number (such as 1, 2 or 3, etc., not specifically limited here) of target lift data are filtered according to the values from large to small, such as the lift value corresponding to bin number 1, or the lift values corresponding to bin number 1 to bin number 2, or the lift values corresponding to bin number 1 to bin number 3.
[0231] Further, when screening a preset number of target lift data of 1, the target lift data corresponding to each group of test result data, that is, the lift value corresponding to the bin number 1, is used as the test lift data corresponding to each group of test result data; similarly, when screening a preset number of target lift data of 2, the sum of the target lift data corresponding to each group of test result data, that is, the sum of the lift values corresponding to the bin number 1 to the bin number 2, is used as the test lift data corresponding to each group of test result data; similarly, when screening a preset number of target lift data of 3, the sum of the target lift data corresponding to each group of test result data, that is, the sum of the lift values corresponding to the bin number 1 to the bin number 3, is used as the test lift data corresponding to each group of test result data.
[0232] The following is an introduction to the application method of the classification prediction model in this application. Fig.13 , an embodiment of the application method of the classification prediction model in the embodiment of the present application includes:
[0233] In step S1301, the user information to be predicted is input into the classification prediction model, and the classification prediction model outputs the prediction probability value;
[0234] In step S1302, if the predicted probability value is greater than the preset probability threshold, the product information push operation is performed to the predicted user;
[0235] In step S1303, if the predicted probability value is less than or equal to the preset probability threshold, the product information push operation is not performed.
[0236] In an embodiment of the present application, after obtaining a trained classification prediction model, the prediction probability value can be output by inputting the user information to be predicted into the classification prediction model. If the prediction probability value is greater than a preset probability threshold, the predicted user corresponding to the prediction probability value is the target object, that is, a product information push operation can be performed to the predicted user. Conversely, if the prediction probability value is less than or equal to the preset probability threshold, the product information push operation is not performed.
[0237] Specifically, after obtaining the trained classification prediction model, feature data (object nickname, object product type preference, etc.) corresponding to the type of the above-mentioned target feature data can be firstly screened out from the user's attribute information as the user information to be predicted.
[0238] Furthermore, based on inputting the user information to be predicted into the classification prediction model, the classification prediction model can output a predicted probability value (between 0 and 1). Then, the predicted probability value (between 0 and 1) can be compared with a preset probability threshold (for example, usually set to 0.5, without specific limitation here). If the predicted probability value is greater than the preset probability threshold, the predicted user corresponding to the predicted probability value is the target object, that is, the product information push operation can be performed to the predicted user. Conversely, if the predicted probability value is less than or equal to the preset probability threshold, the product information push operation is not performed.
[0239] Alternatively, this embodiment can also sort the predicted probability values in descending order (i.e., the higher the probability value, the higher the probability that the product purchasing object will apply for the revolving business product within the next month), and filter out the top M (where M is usually set to 10, and there is no specific restriction here) predicted users with high predicted probability values in order, as cross-marketing high potential response product purchasing objects, and perform targeted product information push operations.
[0240] The following is a detailed description of the training device for the classification prediction model in this application. Fig.16 , Fig.16 This is a schematic diagram of an embodiment of a training device for a classification prediction model in an embodiment of the present application. The training device 20 for a classification prediction model includes:
[0241] An acquisition unit 201 is used to acquire original sample data carrying label information, wherein the original sample data includes model training data and model test data;
[0242] The acquisition unit 201 is also used to acquire original variable data corresponding to the model training data;
[0243] The processing unit 202 is used to perform a feature derivation operation on the original variable data to obtain derived feature data;
[0244] The processing unit 202 is further used to perform a preprocessing operation on the derived feature data to obtain preprocessed feature data;
[0245] The processing unit 202 is further used to perform a feature screening operation on the pre-processed feature data to obtain target feature data;
[0246] The processing unit 202 is further used to construct N original classification prediction models according to the model type, and perform model training operations on the original classification prediction models based on the target feature data and the label information corresponding to the target feature data, to obtain N trained classification prediction models, where N is an integer greater than 1;
[0247] The processing unit 202 is further used to obtain test feature data corresponding to the type of the target feature data in the model test data, and input the test feature data into N trained classification prediction models respectively to obtain N groups of test result data;
[0248] The processing unit 202 is further configured to perform lift calculation operations on the N groups of test result data respectively to obtain test lift data corresponding to each group of test result data;
[0249] The determination unit 203 is used to use the trained classification prediction model with the highest test lift data as the target classification prediction model.
[0250] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided in the embodiment of the present application, the acquisition unit 201 can be specifically used for:
[0251] The consumption time interval, consumption frequency and consumption amount of the original variable data are obtained, and the consumption time interval, consumption frequency and consumption amount are used as derived characteristic data.
[0252] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0253] The business meaning derived data of the original variable data is obtained according to the business meaning, and the business meaning derived data is used as the derived feature data.
[0254] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0255] The original variable data is input into the integrated tree model for discretization and crossover, and the feature data obtained after discretization and crossover is used as the derived feature data.
[0256] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0257] Get the missing value ratio of derived feature data;
[0258] The derived feature data whose missing value ratio is greater than a preset missing value ratio threshold are eliminated, and the eliminated derived feature data are used as preprocessing feature data.
[0259] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0260] Get the single value ratio of the derived feature data;
[0261] The derived feature data whose single value ratio is greater than a preset single value ratio threshold are eliminated, and the eliminated derived feature data are used as preprocessing feature data.
[0262] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0263] Get missing values of derived feature data;
[0264] Fill the missing values according to the business meaning, and use the filled derived feature data as preprocessed feature data.
[0265] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0266] Obtain the statistical quantiles of the derived feature data;
[0267] The values of the statistical quantiles greater than the first quantile threshold are replaced by the values corresponding to the first quantile threshold, and the values of the statistical quantiles less than the second quantile threshold are replaced by the values corresponding to the second quantile threshold;
[0268] The replaced derived feature data is used as preprocessing feature data.
[0269] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0270] A data type conversion operation is performed on the derived feature data according to a preset data format, and the derived feature data after the data type conversion is used as preprocessing feature data.
[0271] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0272] Performing a first screening operation on the preprocessed feature data according to the information value to obtain a first screening feature;
[0273] Performing a second screening operation on the first screening feature according to the population stability index to obtain a second screening feature;
[0274] Performing a third screening operation on the second screening feature according to a stepwise regression method to obtain a third screening feature;
[0275] Performing a fourth screening operation on the third screening feature according to the sequence feature selection method to obtain a fourth screening feature;
[0276] A fifth screening operation is performed on the fourth screening feature according to the symbol consistency to obtain target feature data.
[0277] Optionally, in the above Fig.16 On the basis of the corresponding embodiment, in another embodiment of the application device of the classification prediction model provided by the embodiment of the present application, the processing unit 202 can be specifically used for:
[0278] Arrange N groups of test result data in descending order to obtain N groups of test sorting results;
[0279] Divide each group of test sorting results into equal parts according to the preset number of equal parts to obtain N groups of test binning results;
[0280] Calculate the bin lift data of N groups of test bin results respectively;
[0281] In the binned lift data corresponding to the same set of test result data, a preset number of target lift data are filtered according to the values from large to small;
[0282] The sum of the target lift data corresponding to each group of test result data is used as the test lift data corresponding to each group of test result data.
[0283] The following is a detailed description of the application device of the classification prediction model in this application. Fig.17 , Fig.17 This is a schematic diagram of an embodiment of an application device of a classification prediction model in an embodiment of the present application. The application device 30 of the classification prediction model includes:
[0284] The acquisition unit 301 is used to input the user information to be predicted into the classification prediction model, and output the prediction probability value through the classification prediction model;
[0285] The processing unit 302 is configured to push product information to the predicted user if the predicted probability value is greater than a preset probability threshold;
[0286] The processing unit 302 is further configured to not execute the product information push operation if the predicted probability value is less than or equal to a preset probability threshold.
[0287] On the other hand, the present application provides another schematic diagram of a computer device, such as Fig.18 As shown, Fig.18 3 is a schematic diagram of a computer device structure provided in an embodiment of the present application. The computer device 300 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage devices) storing application programs 331 or data 332. Among them, the memory 320 and the storage medium 330 may be temporary storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the computer device 300. Furthermore, the central processing unit 310 may be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium 330 on the computer device 300.
[0288] The computer device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 333, such as Windows Server 2000, Windows Server 2003, Windows Server 2003E, Windows Server 2003R ... TM , Mac OS X TM , Unix TM ,Linux TM , FreeBSD TM etc.
[0289] The computer device 300 is also used to execute the following steps: Figures 2 to 12 The steps in the corresponding embodiments, and Fig.13 The steps in the corresponding embodiments.
[0290] Another aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following Figures 2 to 12 The steps in the corresponding embodiments, and Fig.13 The steps in the corresponding embodiments.
[0291] Another aspect of the present application provides a computer program product comprising a computer program, which, when executed by a processor, implements the following Figures 2 to 12 The steps in the corresponding embodiments, and Fig.13 The steps in the corresponding embodiments.
[0292] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0293] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0294] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0295] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0296] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.
Claims
1. A training method for a classification prediction model, characterized in that: include: Acquire original sample data carrying label information, wherein the original sample data includes model training data and model test data; Obtaining original variable data corresponding to the model training data; Performing a feature derivation operation on the original variable data to obtain derived feature data; Performing a preprocessing operation on the derived feature data to obtain preprocessed feature data; Performing a feature screening operation on the preprocessed feature data to obtain target feature data; Constructing N original classification prediction models according to the model type, and performing model training operations on the original classification prediction models based on the target feature data and the label information corresponding to the target feature data, respectively, to obtain N trained classification prediction models, wherein N is an integer greater than 1; Acquire test feature data corresponding to the type of the target feature data in the model test data, and input the test feature data into the N trained classification prediction models respectively to obtain N groups of test result data; Performing lift calculation operations on the N groups of test result data respectively to obtain test lift data corresponding to each group of test result data; The trained classification prediction model with the highest test lift data is used as the target classification prediction model.
2. The method according to claim 1, characterized in that The performing a feature derivation operation on the original variable data to obtain derived feature data includes: The consumption time interval, consumption frequency and consumption amount of the original variable data are obtained, and the consumption time interval, consumption frequency and consumption amount are used as the derived characteristic data.
3. The method according to claim 1, characterized in that The performing a feature derivation operation on the original variable data to obtain derived feature data includes: The business meaning derived data of the original variable data is obtained according to the business meaning, and the business meaning derived data is used as the derived feature data.
4. The method according to claim 1, characterized in that: The performing a feature derivation operation on the original variable data to obtain derived feature data includes: The original variable data is input into the integrated tree model for discretization and crossover, and the feature data obtained after discretization and crossover is used as the derived feature data.
5. The method according to claim 1, characterized in that The preprocessing operation is performed on the derived features to obtain preprocessed feature data, including: Obtaining the missing value ratio of the derived feature data; The derived feature data whose missing value ratio is greater than a preset missing value ratio threshold are eliminated, and the eliminated derived feature data are used as the preprocessed feature data.
6. The method according to claim 1, characterized in that The preprocessing operation is performed on the derived features to obtain preprocessed feature data, including: obtaining a single value ratio of the derived feature data; The derived feature data whose single value ratio is greater than a preset single value ratio threshold are eliminated, and the eliminated derived feature data are used as the preprocessed feature data.
7. The method according to claim 1, characterized in that The preprocessing operation is performed on the derived features to obtain preprocessed feature data, including: Obtaining missing values of the derived feature data; The missing values are filled in according to the business meaning, and the derived feature data after filling is used as the preprocessed feature data.
8. The method according to claim 1, characterized in that The preprocessing operation is performed on the derived features to obtain preprocessed feature data, including: Obtaining the statistical quantile of the derived feature data; The values of the statistical quantiles greater than the first quantile threshold are replaced by the values corresponding to the first quantile threshold, and the values of the statistical quantiles less than the second quantile threshold are replaced by the values corresponding to the second quantile threshold; The replaced derived feature data is used as the pre-processed feature data.
9. The method according to claim 1, characterized in that: The preprocessing operation is performed on the derived features to obtain preprocessed feature data, including: A data type conversion operation is performed on the derived feature data according to a preset data format, and the derived feature data after the data type conversion is used as the preprocessing feature data.
10. The method according to claim 1, characterized in that The performing a feature screening operation on the preprocessed feature data to obtain target feature data includes: Performing a first screening operation on the preprocessed feature data according to the information value to obtain a first screening feature; Performing a second screening operation on the first screening feature according to the population stability index to obtain a second screening feature; Performing a third screening operation on the second screening feature according to a stepwise regression method to obtain a third screening feature; Performing a fourth screening operation on the third screening feature according to a sequence feature selection method to obtain a fourth screening feature; A fifth screening operation is performed on the fourth screening feature according to symbol consistency to obtain the target feature data.
11. The method according to claim 1, characterized in that: The performing lift calculation operations on the N groups of test result data respectively to obtain test lift data corresponding to each group of test result data includes: Arrange the N groups of test result data in descending order to obtain N groups of test sorting results; Divide each group of test sorting results into equal parts according to the preset number of equal parts to obtain N groups of test binning results; Calculate the binning lift data of the N groups of test binning results respectively; In the binned lift data corresponding to the same set of test result data, a preset number of target lift data are filtered according to the values from large to small; The sum of the target lift data corresponding to each group of test result data is used as the test lift data corresponding to each group of test result data.
12. An application method of a classification prediction model, characterized in that: include: Inputting the user information to be predicted into the classification prediction model described in any one of claims 1 to 11, and outputting the prediction probability value through the classification prediction model; If the predicted probability value is greater than a preset probability threshold, a product information push operation is performed to the predicted user; If the predicted probability value is less than or equal to the preset probability threshold, the product information push operation is not performed.
13. A training device for a classification prediction model, characterized in that: include: An acquisition unit, used to acquire original sample data carrying label information, wherein the original sample data includes model training data and model test data; The acquisition unit is further used to acquire original variable data corresponding to the model training data; A processing unit, used for performing a feature derivation operation on the original variable data to obtain derived feature data; The processing unit is further used to perform a preprocessing operation on the derived feature data to obtain preprocessed feature data; The processing unit is further used to perform a feature screening operation on the pre-processed feature data to obtain target feature data; The processing unit is further used to construct N original classification prediction models according to the model type, and perform model training operations on the original classification prediction models based on the target feature data and the label information corresponding to the target feature data, to obtain N trained classification prediction models, wherein N is an integer greater than 1; The processing unit is further used to obtain test feature data corresponding to the type of the target feature data in the model test data, and input the test feature data into the N trained classification prediction models respectively to obtain N groups of test result data; The processing unit is further configured to perform lift calculation operations on the N groups of test result data respectively to obtain test lift data corresponding to each group of test result data; A determination unit is used to use the trained classification prediction model with the highest test lift data as the target classification prediction model.
14. An application device of a classification prediction model, characterized in that: include: An acquisition unit, used for inputting the user information to be predicted into the classification prediction model, and outputting the prediction probability value through the classification prediction model; A processing unit, configured to execute a product information push operation to the predicted user if the predicted probability value is greater than a preset probability threshold; The processing unit is further configured to not execute the product information push operation if the predicted probability value is less than or equal to the preset probability threshold.
15. A computer device comprising a memory, a processor and a bus system, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 11 and the steps of the method according to claim 12 are implemented; The bus system is used to connect the memory and the processor so that the memory and the processor can communicate with each other.
16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 and the steps of the method according to claim 12 are implemented.
17. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 11 and the steps of the method according to claim 12.
Citation Information
Cited By
Bladder urination tracking training system based on ultrasonic detection
CN120339243A
