A method and device for sample generation and survival assessment based on data genetic variation

Abstract

The present invention discloses a method and device for sample generation and survival assessment based on data genetic variation, belonging to the field of computer technology. The technical solution includes: parent sample preparation, parent sample crossover, feature inheritance, feature variation, first-level static survival assessment, second-level static survival assessment, dynamic survival assessment, and business modeling; the above steps are repeated to generate a child sample set S2, and static survival assessment and dynamic survival assessment are performed on the child samples in the child sample set S2 to determine whether to eliminate or retain. The purpose is to obtain samples sufficient for modeling in the early stages of the business, making the model establishment more effective, which is more conducive to reducing business risks and improving the profitability of financial institutions.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a method and device for sample generation and survival assessment based on data genetic variation. Background Art

[0002] Machine learning is currently used in a wide range of scenarios, including finance, communications, healthcare, transportation, e-commerce, and more. Many new businesses experience a cold start phase during which it is difficult to apply machine learning methods to build models due to a lack of samples or only a small number of samples.

[0003] The most typical example is how financial institutions create a scoring model to distinguish good and bad customers based on a small amount of existing customer information at the beginning of the target scenario business. For this situation, the general methods currently used are:

[0004] 1. Expanding samples of similar businesses. For example, if the goal is to build a scoring model for auto loans, but the sample size is insufficient, samples from other similar businesses, such as consumer loans, can be used as supplementary data to expand the original auto loan sample size. This approach can alleviate the sample shortage issue, but determining similar businesses and selecting samples often requires manual rules, which introduces uncertainty. Often, the actual feature space distribution of "similar customer groups" differs significantly from that of the target business customer group, making the results unpredictable.

[0005] 2. Sample sampling, such as random oversampling and random downsampling, deals with the problem of sample imbalance. In essence, it does not add new samples, but instead copies or reduces samples through sampling to balance the positive and negative ratios of training samples.

[0006] 3. The SMOTE method works as follows: select the feature space of a minority class, find its k nearest neighbors, select one of these neighbors, and place a synthetic point anywhere on the line connecting the considered point and its selected neighbor. Synthetic points are not inherently completely outside the feature space of the minority class samples. If the feature space of the minority class samples undergoes significant changes over time, the effectiveness of SMOTE will generally be compromised.

[0007] 4. Rejection inference: This is a common sample expansion scenario in the financial lending sector. When a customer's loan application is rejected, the financial institution obtains the customer's characteristic information but cannot predict their future label performance. Therefore, rejection inference is needed to reasonably infer the customer's label. Numerous rejection inference methods exist, each with varying effectiveness depending on the scenario, and there is no universally recognized effective solution. Summary of the Invention

[0008] In order to solve the problem that the existing technology has a lack of samples or only a small number of samples, making it difficult to apply machine learning methods to build models, the present invention provides a sample generation and survival assessment method and device based on data genetic variation. Its purpose is to obtain samples sufficient for modeling in the early stages of the business, so that the model building effect is better, which is more conducive to reducing business risks and improving the profitability of financial institutions.

[0009] The technical solution adopted in the present invention is as follows:

[0010] A method for sample generation and survival assessment based on genetic variation in data, specifically comprising:

[0011] Parent sample preparation: Obtain the initial parent sample based on the business scenario;

[0012] Parent sample crossover: combine various initial parent samples and cross-obtain labels for offspring samples;

[0013] Characteristic inheritance: Determine the characteristics of the offspring sample, compare the characteristic vector Xi to a DNA chain, set the inheritance coefficient H, randomly select H*i characteristic genes from the characteristic space Xi of the two parent samples, exchange their values, and generate two new offspring characteristic vectors Zi. The characteristic vector length of the offspring sample is consistent with that of the parent sample;

[0014] Feature variation: Get the feature vector Zi of the offspring sample, set the variation coefficient V according to different features, and transform the offspring feature vector Zi exchanged in the feature inheritance step according to the variation coefficient V to obtain the final feature vector Zi' of the offspring sample and the offspring sample set S1;

[0015] Static survival evaluation: Through different crossover and mutation combinations, a large number of offspring sample sets S1 are obtained. Static survival evaluation is performed on the offspring samples. A multi-classifier evaluation voting method is used to perform static production evaluation. The classifier pass rate is calculated by combining the voting results of multiple classifiers. Offspring samples with low pass rates are eliminated, and the offspring sample set that has passed is retained as S1';

[0016] Dynamic survival evaluation: For the offspring samples that have been retained after the static survival evaluation, a multi-classifier evaluation voting method is used to perform dynamic production evaluation in multiple time windows. The pass rate of each time window is calculated by combining the voting results of multiple classifiers. The pass rate of each time window of the offspring samples is evaluated in turn. The offspring samples that do not meet the survival round are eliminated, and the offspring sample set that has passed the survival round becomes S1.

[0017] Business modeling: During the business modeling phase, for the offspring samples that survive two rounds of survival assessment, survival rounds and pass rate thresholds are determined based on specific scenarios. Regular samples are then added to the modeling samples for business modeling.

[0018] The above steps are repeated from parent sample preparation to business modeling to generate a child sample set S2, and the child samples in the child sample set S2 are subjected to the static survival evaluation and dynamic survival evaluation to determine whether to eliminate or retain.

[0019] By adopting the above technical solution, the setting of the genetic coefficient can greatly expand the randomness of the feature intersection of the offspring samples, thereby enriching the generation method of the offspring samples; by setting the feature variation and the variation coefficient, the randomness and diversity of the features of the offspring samples can be further expanded, further enriching the generation method of the offspring samples; through the idea of ​​genetic variation, some similar real samples are simulated, and the business model can be established in combination with the obtained real samples, and the model learns a more comprehensive potential distribution, which is of great help for future business expansion; all samples are generated using the data of the business itself, without relying on other similar businesses, solving the problem of no similar businesses or similar businesses that cannot be accurately defined; the generated offspring samples are more consistent with the distribution of the original parent customer group, and are more effective for the establishment of subsequent business models; not only is data processing performed from the algorithm, but also the prior experience of the business can be incorporated, the generated samples are more diverse and reasonable, and through static and dynamic survival evaluation, the randomness of sample generation is also limited, which is more reasonable and effective than the pure algorithm construction method.

[0020] Optionally, the setting of the coefficient of variation must meet the following two basic principles:

[0021] 1) Range rationality principle: For a certain proportional feature, if the original value is 0 to 1, the value of the mutated feature must also be 0 to 1;

[0022] 2) Distribution preservation principle: For characteristics that conform to the normal distribution, the standard deviation of the parent sample characteristics is calculated to be σ, the mean value is μ, and the probability that the characteristic value falls within the interval of μ±3σ is 99.73%. Then, after the offspring characteristics mutate, the value range should be between μ±3σ to ensure that the characteristic values ​​of the entire sample population do not deviate from the normal distribution.

[0023] Based on the above, by setting the feature variation and coefficient of variation, the randomness and diversity of the features of the offspring samples can be further expanded, and the generation method of the offspring samples can be further enriched.

[0024] Optionally, the static survival assessment is divided into two stages, specifically:

[0025] One-stage static survival assessment: This is to set combination rules based on prior experience of feature vectors to eliminate some samples with obvious defects;

[0026] Second-stage static survival evaluation: After the first-stage survival evaluation, the remaining offspring samples need to pass the model evaluation. The classifier pass rate is calculated by combining the voting results of multiple classifiers, and an elimination threshold is set. The offspring sample groups with low pass rates are eliminated, and the passing sample groups and the corresponding classifier pass rates are retained. The classifier pass rate can be used as the T1 stage survival weight of the sample group. After the T1 stage survival evaluation, the remaining offspring sample set is S1'.

[0027] Optionally, the model evaluation method is: first divide the D1 parent sample into a modeling data set train1 and a test data set test1, use train1 to train multiple classifiers, evaluate the effect of the trained classifiers on the test set test1, and use the commonly used binary classification evaluation indicator AUC as the evaluation indicator. The obtained AUC results of multiple classifiers are used as a benchmark, and the S1 offspring samples are equally divided into M groups. After each group is merged with the train1 sample, multiple classifiers are also trained, and the AUC is calculated on test1 respectively. Finally, the AUC before and after are compared on multiple classifiers. If the AUC is improved after adding the offspring, then the classifier votes 1, otherwise it votes 0. The classifier pass rate is calculated by combining the voting results of multiple classifiers, and an elimination threshold is set to eliminate the offspring sample group with a low pass rate, retaining the passing sample group and the corresponding classifier pass rate. After the T1 stage survival evaluation, the remaining offspring sample set is S1'.

[0028] Optionally, the classifier pass rate is used as the survival weight of the offspring samples passing the T2 stage, and the survival weight is used as the sample weight for modeling.

[0029] In addition, to achieve the above-mentioned purpose, the present invention also provides a sample generation and survival assessment device based on data genetic variation, the device comprising:

[0030] Business data storage module: used to receive and store business data and provide initial parent samples;

[0031] Massive sample generation module: used to receive the initial parent sample and generate massive sub-generation samples in accordance with the parameters and rules provided by the rule configuration module;

[0032] Rule configuration module: used for visual configuration of genetic coefficient, variation coefficient, and static survival rule;

[0033] Survival model operation module: used to train and deploy multiple survival model classifiers, perform static and dynamic generation evaluations on offspring samples, and output sample survival results and corresponding survival weights;

[0034] Sample retention module: used to save samples that pass the survival assessment and the corresponding survival weights;

[0035] Business model operation module: a module used to integrate offspring samples and parent samples, perform modeling, and deploy applications. The model results are applied to the business.

[0036] Based on the above device, the business channel obtains business-related data and stores it in the business data storage module. The business data storage module is responsible for providing parent sample data. The rule configuration module configures the genetic coefficient, coefficient of variation, and static survival rule configuration. The survival model operation module deploys multiple classification models for static and dynamic survival model evaluation. The business data storage module, rule configuration module, and survival model operation module work together, and retained samples enter the retained sample data storage module. The sample retention module and the business data storage module jointly provide samples to establish a business model, which is deployed in the business model operation module. The business model is finally applied to the business channel to affect the business.

[0037] In addition, to achieve the above-mentioned object, the present invention further provides a computer device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0038] Memory for storing computer programs;

[0039] The processor is configured to implement the above-mentioned sample generation and survival assessment method based on data genetic variation when executing the program stored in the memory.

[0040] In addition, to achieve the above-mentioned purpose, the present invention also provides a storage medium, which stores instructions. When the storage medium is run on a computer, it enables the computer to execute the above-mentioned sample generation and survival assessment method based on data genetic variation.

[0041] In addition, to achieve the above objectives, the present invention also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the above-mentioned sample generation and survival assessment method based on data genetic variation.

[0042] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0043] 1) Unlike the existing "similar business sample expansion" solution, this solution uses data from the business itself to generate samples, independent of other similar businesses. This solves the problem of similar businesses not existing or difficult to accurately define. The generated offspring samples are more consistent with the distribution of the original parent customer base, making them more effective for building subsequent business models.

[0044] 2) Unlike existing "sample sampling" techniques, this solution doesn't simply replicate or reduce the original parental sample. Instead, it leverages the concept of genetic variation to create a completely new sample, generating offspring samples that are related to but distinct from the original parental sample. Due to business expansion limitations, it's often impossible to obtain all available real samples. However, this solution leverages the concept of genetic variation to simulate some similar real samples. These samples can then be combined to build a business model. This model learns a more comprehensive underlying distribution, significantly aiding future business expansion.

[0045] 3) Unlike the existing SMOTE method, the sample generation method in this technical solution not only processes data algorithms but also incorporates prior business experience, resulting in more diverse and reasonable samples. Furthermore, through static and dynamic survival assessments, the randomness of sample generation is also limited. By combining algorithms, expert experience, and survival assessments, the entire solution forms a closed loop. The resulting offspring samples are more reasonable and effective than those generated purely by algorithms.

[0046] 4) Unlike the existing "rejection inference" technique, this technique uses rejected samples as parent samples, interleaving them with positive and negative samples with known labels to generate new offspring samples with known labels. Rejection inference, on the other hand, infers labels for parent samples with unknown labels. This approach is completely different and can serve as a new type of rejection inference method. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0048] Figure 1 This is a step flow of a sample generation and survival assessment method based on data genetic variation in the present invention. Figure 1 ;

[0049] Figure 2 This is a process of a sample generation and survival assessment method based on data genetic variation provided in Example 1 Figure 2 ;

[0050] Figure 3 This is a structural diagram of a sample generation and survival assessment device based on data genetic variation provided in Example 2;

[0051] Figure 4 This is a structural diagram of a computer device provided in Example 3. DETAILED DESCRIPTION

[0052] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0053] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus 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 are inherent to these processes, methods, products or apparatuses.

[0055] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0056] Example 1:

[0057] like Figure 1 As shown, the present invention provides a method for sample generation and survival assessment based on genetic variation data, which may include the following steps:

[0058] S101. Parent Sample Preparation: Based on the business scenario, obtain the initial parent samples. The time window of the initial parent samples is T1, and the feature space is Xi (i is the dimension of the feature vector, usually a high-dimensional vector). The labels of the initial parent samples are divided into three categories: positive, negative, and uncertain. The total number of initial parent samples is D1, of which positive samples are Dp, negative samples are Dn, and uncertain samples are Du. Obviously, Dp + Dn + Du = D1.

[0059] S102, Parent Crossover: Combine various initial parent samples and crossover to obtain labels for offspring samples. There are many specific crossover schemes, such as positive-positive crossover to obtain positive labels, positive-uncertain crossover to obtain positive labels, negative-uncertain crossover to obtain negative labels, and negative-negative crossover to obtain negative labels.

[0060] S103. Feature inheritance: After the combination and label generation in step S102, the next step is to determine the characteristics of the offspring samples. Similar to the idea of ​​DNA gene crossover, the feature vector Xi is regarded as a DNA chain, and the genetic coefficient H (value 0-1) is set. The feature space Xi of the two parent samples randomly selects H*i feature genes to exchange values ​​to generate two new offspring sample feature vectors Zi. The feature vector length of the offspring sample is consistent with that of the parent. The feature vector is compared to the DNA chain. By setting the genetic coefficient, the randomness of the feature crossover of the offspring samples can be greatly expanded, thereby enriching the generation method of the offspring samples.

[0061] S104, Feature Mutation: Through step S103, the feature vector Zi of the offspring sample is obtained. The coefficient of variation V is set according to different features. The feature vector Zi of the offspring sample exchanged in step S103 is transformed according to the coefficient of variation V to obtain the final feature vector Zi' of the offspring sample. The offspring sample set is S1. The method for setting the coefficient of variation requires prior experience. To ensure the rationality of the values ​​after mutation, the following two basic principles must be met: 1. The principle of range rationality. For example, for a certain proportional feature, if the original value is 0 to 1, the value of the feature after mutation must also be 0 to 1; 2. The principle of distribution preservation. For example, for certain features with clear meanings, such as human height, there is prior knowledge that it is normally distributed. According to the theory of normal distribution, the standard deviation of the parent sample height is calculated to be σ, and the mean is μ. The probability of a height value falling within the range of μ ± 3σ is known to be 99.73%. Therefore, the range of the offspring height after mutation should be between μ ± 3σ, ensuring that the height feature of the entire sample population does not deviate from the normal distribution. By setting the feature variation and coefficient of variation, the randomness and diversity of the features of the offspring samples can be further expanded, and the generation method of the offspring samples can be further enriched.

[0062] S105. One-stage static survival evaluation: According to steps S102-S104, a large number of offspring sample sets S1 can be obtained through different crossover mutation combinations. The offspring samples need to be evaluated for survival to determine whether to eliminate or retain them. The static survival evaluation is divided into two stages. The first stage is rule evaluation, that is, setting some combination rules based on the prior experience of feature vectors to exclude some obviously defective samples. For example, if there are two features of age + degree in the features, then if there are samples of 3 years old + doctorate in the offspring samples, they need to be directly eliminated.

[0063] S106, Second-Stage Static Survival Evaluation: After the first-stage static survival evaluation, the remaining offspring samples undergo model evaluation. The model evaluation method is as follows: First, divide the parent samples D1 into a modeling dataset, train1, and a test dataset, test1. Use train1 to train multiple classifiers, using models such as linear models, logistic regression, or complex models like LightGBM. The trained classifiers are evaluated on the test set, test1, using the commonly used binary classification metric, AUC. The AUC results of the multiple classifiers are used as a benchmark. Next, divide the offspring samples in the offspring sample set S1 into M groups. Each group is then merged with the train1 samples and trained with multiple classifiers. The AUCs are then calculated on test1. Finally, the AUCs are compared on multiple classifiers. If the AUC improves after adding the offspring, the classifier is assigned a vote of 1; otherwise, it is assigned a vote of 0. The voting results of multiple classifiers are combined to calculate the classifier pass rate, which is the average of the classifier voting results. A elimination threshold is set to eliminate offspring sample groups with low pass rates, retaining the passing sample groups and their corresponding classifier pass rates. The classifier pass rates serve as the T1 survival weights for these sample groups. The remaining offspring sample set after the T1 survival assessment is designated S1'. A two-stage static survival assessment utilizes a multi-classifier voting method to ensure the offspring sample's adaptability to various model algorithms during subsequent business modeling. Survival weights are set to describe the degree of similarity between offspring samples and their parent samples, and to weight the offspring sample model training in subsequent business modeling.

[0064] S107. Dynamic survival evaluation: After two stages of static survival evaluation, the adaptability of the offspring samples in the remaining offspring sample set S1' to the T1 time has been effectively evaluated. As the business continues to develop, time rolls to the T2 window, and new parent samples D2 appear. Referring to step 6, the D2 samples are also divided into the training set train2 and the test set test2. Similarly, multiple classifiers are trained using train2, and the AUC of each classifier is obtained on test2. The offspring samples in the offspring sample set S1' are merged with the train2 samples according to the previous grouping and then trained on the same multiple classifiers. The AUC is calculated on test2 respectively. Finally, the AUC before and after are compared on multiple classifiers. If the AUC is improved after adding the offspring, the classifier votes 1, otherwise it votes 0. The classifier pass rate is calculated based on the voting results of multiple classifiers, and samples with low pass rates are eliminated. In addition, the passing samples and the corresponding classifier pass rates are retained. The classifier pass rate can be used as the T2 stage survival weight of the passing samples. After the survival evaluation in stage T2, the remaining offspring sample set of T1 becomes S1. As the business evolves over time, the distribution of customer groups will change. Setting up a dynamic survival evaluation solves the problem that offspring samples can only adapt to the business environment of a certain period. By adding these offspring samples that have passed the test, the business model will have better generalization performance.

[0065] S108, Business Modeling: Repeat step S107 to dynamically evaluate the offspring samples in offspring sample set S1 over multiple time windows. The more rounds they pass, the stronger the offspring sample's survival ability. During the business modeling phase, flexible survival rounds and pass rate thresholds are set, and samples that pass the survival assessment are added to the modeling sample set for business modeling. Survival weights can also be used as sample weights for modeling. While the actual sample survival weight is naturally 1, different sample weights can also affect model training results.

[0066] In addition to performing dynamic survival evaluation on the offspring samples in the offspring sample set S1, the parent sample D2 in the S109 and T2 stages can also generate the S2 offspring sample set according to steps S101-S107, and perform static evaluation at T2, as well as dynamic evaluation in subsequent T3, T4... to determine elimination or retention.

[0067] The sample generation and survival assessment method proposed in this embodiment can be applied to the following two scenarios:

[0068] Scenario 1: Business Cold Start. For example, in the financial sector, when a credit business is first launched, risk control typically relies solely on expert experience or rules derived from analyzing a small amount of data. Machine learning methods can only be used to build models when a sufficient sample size has been accumulated. Furthermore, borrowers' repayments require a performance period to determine whether they will become overdue, typically lasting from several months to a year. This performance period also prolongs the sample accumulation process, typically requiring one to two years to accumulate sufficient samples for modeling. However, if we can generate approximate samples based on genetic variation, we can obtain sufficient samples for modeling early in the business, which can help reduce business risks and improve the profitability of financial institutions.

[0069] Scenario 2: Anomaly detection scenario. In most companies, there are only a few abnormal cases, such as malicious order swiping, scalper orders, credit card fraud, electricity theft, equipment failure, etc. The proportion of these data samples is usually a small part of the overall sample. Taking credit card fraud as an example, the fraud rate of physical credit cards is generally within 0.1%. If we can generate low-proportion samples through the idea of ​​genetic variation, we can alleviate the sample imbalance problem and make the model more effective. This solves the problem of modeling difficulties in the above two technical backgrounds.

[0070] Example 2:

[0071] Another aspect of the present invention provides a device for implementing the above-mentioned method for sample generation and survival assessment based on genetic variation of data, please refer to Figure 2 , the device comprises:

[0072] Business data storage module: used to receive and store business data and provide initial parent samples;

[0073] Massive sample generation module: used to accept the initial parent sample and generate massive sub-generation samples in accordance with the parameters and rules provided by the rule configuration module;

[0074] Rule configuration module: used for visual configuration of genetic coefficient, variation coefficient, and static survival rule;

[0075] Survival model operation module: used to train and deploy multiple survival model classifiers, to perform static and dynamic generation evaluations on offspring samples, and to output sample survival results and corresponding survival weights;

[0076] Sample retention module: used to save samples that pass the survival assessment and the corresponding survival weights;

[0077] Business model operation module: a module used to integrate offspring samples and parent samples, perform modeling, and deploy applications. The model results are applied to the business.

[0078] Based on the above device, the business channel obtains business-related data and stores it in the business data storage module. The business data storage module is responsible for providing parent sample data. The rule configuration module configures the genetic coefficient, coefficient of variation, and static survival rule configuration. The survival model operation module deploys multiple classification models for static and dynamic survival model evaluation. The business data storage module, rule configuration module, and survival model operation module work together, and retained samples enter the retained sample data storage module. The sample retention module and the business data storage module jointly provide samples to establish a business model, which is deployed in the business model operation module. The business model is finally applied to the business channel to affect the business.

[0079] Example 3:

[0080] Another aspect of the present invention provides a computer device, such as Figure 3 As shown, the computer device includes a memory, a processor, a communication interface and a communication bus. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, the steps in the above embodiment method are implemented.

[0081] The processor may be a central processing unit (CPU).

[0082] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as the corresponding program units in the above-described method embodiments of the present invention. The processor executes the non-transitory software programs, instructions, and modules stored in memory to perform various processor functions and work data processing, thereby implementing the methods in the above-described method embodiments.

[0083] The memory may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory may optionally include a memory remotely located relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0084] The one or more units are stored in the memory, and when executed by the processor, perform the method in the above embodiment.

[0085] The specific details of the above-mentioned computer device can be understood by referring to the corresponding descriptions and effects in the above-mentioned embodiments, and will not be repeated here.

[0086] Example 4:

[0087] On the other hand, the present invention further provides a computer storage medium having instructions stored therein, which, when run on a computer, enables the computer to execute the sample generation and survival assessment method described in any of the above embodiments. Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, which, when executed, may include the processes of the embodiments of the above-mentioned methods. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above-mentioned types of memory.

[0088] Example 5:

[0089] In another embodiment provided by the present invention, a computer program product comprising instructions is also provided, which, when executed on a computer, enables the computer to execute the sample generation and survival assessment method described in any one of the above embodiments.

[0090] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted from one storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0091] Obviously, those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computing device. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0092] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for sample generation and survival assessment based on genetic variation of data, characterized in that: Specifically include: Parent sample preparation: Obtain the initial parent sample based on the business scenario; Parent sample crossover: combine various initial parent samples and cross-obtain labels for offspring samples; Characteristic inheritance: Determine the characteristics of the offspring sample, compare the characteristic vector Xi to a DNA chain, set the inheritance coefficient H, randomly select H*i characteristic genes from the characteristic space Xi of the two parent samples, exchange their values, and generate two new offspring characteristic vectors Zi. The characteristic vector length of the offspring sample is consistent with that of the parent sample; Feature variation: Get the feature vector Zi of the offspring sample, set the variation coefficient V according to different features, and transform the offspring feature vector Zi exchanged in the feature inheritance step according to the variation coefficient V to obtain the final feature vector Zi' of the offspring sample and the offspring sample set S1; Static survival evaluation: Through different crossover and mutation combinations, a large number of offspring sample sets S1 are obtained. Static survival evaluation is performed on the offspring samples. A multi-classifier evaluation voting method is used to perform static production evaluation. The classifier pass rate is calculated by combining the voting results of multiple classifiers. The classifier pass rate is used as the T1 stage survival weight of the sample group. Offspring samples with low pass rates are eliminated, and the offspring sample set that has passed is retained as S1'; Dynamic survival evaluation: For the offspring samples retained after the static survival evaluation, a multi-classifier evaluation voting method is used to perform dynamic production evaluation in multiple time windows. The pass rate of each time window is calculated by combining the voting results of multiple classifiers. The classifier pass rate is used as the T2 stage survival weight of the passing sample. The pass rate of each time window of the offspring samples is evaluated in turn. The offspring samples that do not meet the survival round are eliminated, and the remaining offspring sample set becomes S1''; Business modeling: During the business modeling phase, for the offspring samples that survive two rounds of survival assessment, survival rounds and pass rate thresholds are set based on specific scenarios. Regular samples are added to the modeling samples for business modeling. The survival weight is used as the sample weight for modeling. The actual sample survival weight is naturally 1. The above steps are repeated from parent sample preparation to business modeling to generate a child sample set S2, and the static survival evaluation and dynamic survival evaluation are performed on the child samples in the child sample set S2 to determine whether to eliminate or retain.

2. A method for sample generation and survival assessment based on data genetic variation according to claim 1, characterized in that: The setting of the coefficient of variation V must meet the following two basic principles: 1) Range rationality principle: for a certain proportional feature, if the original value is 0 to 1, the value of the feature after variation must also be 0 to 1; 2) Distribution preservation principle: For characteristics that conform to the normal distribution, the standard deviation of the parent sample characteristics is calculated to be σ, the mean is μ, and the probability that the characteristic value falls within the interval of μ±3σ is known to be 99.73%. Then, the value range of the offspring characteristics after mutation should be between μ±3σ to ensure that the characteristic values ​​of the entire sample population do not deviate from the normal distribution.

3. The method for sample generation and survival assessment based on data genetic variation according to claim 1, characterized in that: The static survival assessment is divided into two stages, specifically: One-stage static survival assessment: This is to set combination rules based on prior experience of feature vectors to eliminate some samples with obvious defects; Second-stage static survival evaluation: After the first-stage survival evaluation, the remaining offspring samples need to pass the model evaluation. The classifier pass rate is calculated by combining the voting results of multiple classifiers, and an elimination threshold is set. The offspring sample groups with low pass rates are eliminated, and the passing sample groups and the corresponding classifier pass rates are retained. The classifier pass rate is used as the T1 stage survival weight of the sample group. After the T1 stage survival evaluation, the remaining offspring sample set is S1'.

4. The method for sample generation and survival assessment based on data genetic variation according to claim 3, characterized in that: The model evaluation method is as follows: first, the D1 parent sample is divided into a modeling data set train1 and a test data set test1, multiple classifiers are trained using train1, and the trained classifiers are evaluated on the test set test1. The commonly used binary classification evaluation indicator AUC is selected as the evaluation indicator, and the AUC results of multiple classifiers are obtained as a benchmark. The S1 offspring sample is equally divided into M groups, and each group is merged with the train1 sample and multiple classifiers are trained in the same way. The AUC is calculated on test1 respectively, and finally the AUC before and after are compared on multiple classifiers. If the AUC is improved after the offspring is added, then the classifier votes 1, otherwise it votes 0. The classifier pass rate is calculated by combining the voting results of multiple classifiers, and an elimination threshold is set to eliminate the offspring sample group with a low pass rate, retaining the passing sample group and the corresponding classifier pass rate. After the T1 stage survival evaluation, the remaining offspring sample set is S1'.

5. The method for sample generation and survival assessment based on data genetic variation according to claim 4, characterized in that: The classifier passing rate is used as the survival weight of the offspring samples passing the T2 stage, and the survival weight is used as the sample weight for modeling.

6. A device for implementing the method for sample generation and survival assessment based on data genetic variation as described in any one of claims 1 to 5, characterized in that: The device comprises: Business data storage module: used to receive and store business data and provide initial parent samples; Massive sample generation module: used to receive the initial parent sample and generate massive sub-generation samples in accordance with the parameters and rules provided by the rule configuration module; Rule configuration module: used for visual configuration of genetic coefficient, variation coefficient, and static survival rule; Survival model operation module: used to train and deploy multiple survival model classifiers, perform static and dynamic generation evaluations on offspring samples, and output sample survival results and corresponding survival weights; Sample retention module: used to save samples that pass the survival assessment and the corresponding survival weights; Business model operation module: a module used to integrate offspring samples and parent samples, perform modeling, and deploy applications. The model results are applied to the business.

7. A computer device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; A processor, configured to implement the method according to any one of claims 1 to 5 when executing a computer program stored in a memory.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images