Application method and generation method of classification prediction model, equipment, medium and product

Abstract

The invention discloses an application method and a generation method of a classification prediction model, equipment, a medium and a product. The method can be applied to the technical field of big data, and comprises the steps of performing feature extraction on collected data, and obtaining a plurality of input features corresponding to a current prediction object; discrete values in the category features are converted into numerical values for representation, and input data of a classification prediction model are generated; inputting the input data into a pre-trained classification prediction model, and obtaining a classification result output by the classification prediction model; wherein the optimal parameter group used for generating the classification prediction model is obtained by performing multi-round training on the initial classification model; training sample data used for multi-round training of the initial classification model is obtained by preprocessing historical data of a plurality of objects and performing discrete value conversion. By adopting the technical scheme, accurate and rapid classification prediction can be realized when the input features comprise the category features and the high-dimensional sparse features with a nonlinear relationship.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data, and in particular to an application method and a generation method of a classification prediction model, a device, a medium and a product. BACKGROUND

[0002] The classification prediction task can divide the prediction object into a pre-defined category according to the input feature data.

[0003] In the prior art, the classification prediction task is generally realized by a logistic regression method or an extreme gradient boosting method. However, the logistic regression method depends on a linear hypothesis to capture the correlation between the features and the target for prediction, and cannot capture the nonlinear relationship to identify the potential rules and features in the data. The extreme gradient boosting method needs to perform additional preprocessing on the category features, which is prone to cause dimension explosion and increase the computational burden.

[0004] In some classification prediction scenarios, the input features may include category features with nonlinear relationships and high-dimensional sparse features. If the logistic regression method or the extreme gradient boosting method is used for classification prediction, the prediction accuracy will be reduced and the computational burden will be increased. The existing method is not suitable for classification prediction in this scenario. SUMMARY

[0005] The present application provides an application method and a generation method of a classification prediction model, a device, a medium and a product, which can realize accurate and rapid classification prediction when the input features include category features with nonlinear relationships and high-dimensional sparse features.

[0006] According to an aspect of the present application, an application method of a classification prediction model is provided, comprising:

[0007] Feature extraction is performed on the collected data to obtain a plurality of input features corresponding to the current prediction object; wherein the input features include category features and high-dimensional sparse features;

[0008] Discrete values in the category features are converted into numerical representations to generate input data of the classification prediction model;

[0009] The input data is input into a pre-trained classification prediction model to obtain a classification result output by the classification prediction model;

[0010] The optimal parameter set for generating the classification prediction model is obtained after multiple rounds of training of an initial classification model. The training sample data for the multiple rounds of training of the initial classification model is obtained by preprocessing and discrete value conversion of the historical data of a plurality of objects.

[0011] According to another aspect of the present application, a generation method of a classification prediction model is provided, comprising:

[0012] Historical data of multiple objects is acquired and preprocessed to generate initial sample data; wherein, the historical data includes multiple initial features of the objects and target values, and the initial features include categorical features and high-dimensional sparse features;

[0013] The discrete values ​​in the initial sample data are converted into numerical representations to generate training sample data;

[0014] Based on the training sample data and the parameter search space, the initial classification model is trained in multiple rounds, and after each round of training, the score value corresponding to the parameter set used in the current round of training is obtained.

[0015] Once the training termination condition is met, the optimal parameter set is determined based on each score value, and the classification prediction model is trained using the optimal parameter set.

[0016] Optionally, historical data of multiple objects can be obtained, and the historical data can be preprocessed to generate initial sample data, including:

[0017] Retrieve historical data for multiple objects and perform duplicate checks on each piece of historical data. When duplicate historical data is detected, perform a deduplication operation.

[0018] Based on the historical data of each object, the missing proportion of each initial feature is detected. When the missing proportion of the target initial feature is detected to be greater than the proportion threshold, the target initial feature is deleted from each historical data to obtain the initial sample data of each object.

[0019] The advantages of this setup are as follows: duplicate data can cause the model to overlearn on duplicate samples, leading to prediction bias. By deleting duplicate samples, redundant samples can be removed, ensuring that the data is representative. At the same time, initial features with a high missing rate have low feature information value. Therefore, retaining only initial features with certain information value can effectively improve the training efficiency of the model and effectively reduce the collection of redundant features during application.

[0020] Optionally, the discrete values ​​in the initial sample data are converted into numerical representations to generate training sample data, including:

[0021] Obtain the original index sequence of each initial sample data, and reorder the original index sequence using a random algorithm to generate a random index sequence;

[0022] Based on the random index sequence, starting from the first initial sample data, each initial sample data is traversed, and the discrete values ​​of categorical features in each initial sample data are converted into numerical representations in turn.

[0023] The advantages of this setup are: by introducing a random permutation order to simulate time, it ensures that the calculation encoding relies only on historical data, avoiding overfitting caused by the model seeing its own label in advance, thus completely solving the target leakage problem and significantly improving generalization ability. By compressing the categorical features into a numerical feature, the association between the category and the target can be reflected in real time through encoding, while maintaining randomness and unbiasedness.

[0024] Optionally, based on the random index sequence, starting from the first initial sample data, traverse each initial sample data, and sequentially convert the discrete values ​​of categorical features in each initial sample data into numerical representations, including:

[0025] When traversing the initial target sample data, determine the target value of the target categorical feature in the initial target sample data; where the target value is a discrete value.

[0026] Based on the target value of the target category feature, determine the first statistical value and the second statistical value respectively, and calculate the encoding value of the target category feature based on the first statistical value and the second statistical value. Then, replace the target value with the calculated encoding value.

[0027] Based on the target value of the target category feature and the target value of the initial sample data, the first and second statistical values ​​are updated for use when performing discrete value transformation on the next initial sample data.

[0028] The advantage of this setup is that by calculating the encoding in the above way, we can avoid the problem of feature dimension explosion caused by one-hot encoding in existing technologies when there are a large number of categorical features. It can rely only on historical data during the encoding process, solve the target leakage problem, and thus avoid overfitting during model training and improve the model's generalization ability.

[0029] Optionally, based on the training sample data and the parameter search space, the initial classification model is trained in multiple rounds, and after each round of training, a score value corresponding to the parameter set used in the current round of training is obtained, including:

[0030] Before each round of training begins, the current parameter group is selected in the parameter search space according to the parameter sampling strategy, and the training sample data is divided into multiple data groups.

[0031] In each set of data, one set is selected as the validation set and the remaining sets are used as the training set. Based on the current parameter set and the selected validation and training sets, multiple training sessions are conducted, and a single score value is obtained for each training session.

[0032] Calculate the average score of each individual test in the current training round as the score for the current parameter group, and update the parameter sampling strategy based on the score for the current parameter group so that the parameter group can be selected in the next training round according to the updated parameter sampling strategy.

[0033] The advantage of this setup is that cross-validation can reduce the impact of randomness in sample splitting, especially in scenarios where the proportion of a single target value is low, such as the detection of non-conforming products. It avoids the extreme case where no samples are lost in the validation set due to a single split. By calculating the average score, the random fluctuations of a single validation can be offset, and the performance of the parameter group can be evaluated more objectively.

[0034] Optionally, upon determining that the training termination condition is met, the optimal parameter set is determined based on each score value, including:

[0035] When the changes in the score values ​​of multiple consecutive parameter groups are all within the target interval or the maximum number of iterations is reached, the training termination condition is determined to be met.

[0036] Based on the magnitude of each score, the parameter group is divided into a high group and a low group. Based on the actual parameter values ​​of each parameter in the high group and the low group, the optimal value of each parameter is determined, and the optimal values ​​of each parameter are combined to form the optimal parameter group.

[0037] The advantage of this setting is that by determining that the training termination condition is met when the changes in the scores of multiple consecutive parameter groups are all within the target range or the maximum number of iterations is reached, the training efficiency and performance can be balanced through the termination condition, avoiding invalid iterations or overtraining. Thus, training can be terminated even if it is not necessary to save training resources. This ensures the statistical significance of the parameters and that the parameters can be adapted to the specific prediction task. Ultimately, the trained model can capture the influence of potential patterns and features in the data while taking into account generalization ability.

[0038] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0039] At least one processor; and

[0040] A memory communicatively connected to the at least one processor; wherein,

[0041] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to execute the application method of the classification prediction model and the generation method of the classification prediction model according to any embodiment of the present invention.

[0042] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the application method of the classification prediction model and the generation method of the classification prediction model as described in any embodiment of the present invention.

[0043] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the application method of the classification prediction model and the generation method of the classification prediction model as described in any embodiment of the present invention.

[0044] The technical solution of this invention extracts features from the collected data to obtain multiple input features corresponding to the current prediction object, converts the discrete values ​​in the categorical features into numerical representations, generates input data for the classification prediction model, and inputs the input data into the pre-trained classification prediction model to obtain the classification result output by the classification prediction model. This method enables accurate classification prediction of input features containing categorical features and high-dimensional sparse features, effectively reducing the computational load of feature processing during the prediction process, thereby improving classification efficiency.

[0045] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of an application method for a classification prediction model provided in Embodiment 1 of the present invention;

[0048] Figure 2 This is a flowchart of a method for generating a classification prediction model according to Embodiment 2 of the present invention;

[0049] Figure 3 This is a flowchart of another method for generating a classification prediction model according to Embodiment 3 of the present invention;

[0050] Figure 4 This is a schematic diagram of the structure of an application device for a classification prediction model provided in Embodiment 4 of the present invention;

[0051] Figure 5 This is a schematic diagram of the structure of a classification prediction model generation device provided in Embodiment 5 of the present invention;

[0052] Figure 6 This is a schematic diagram of the structure of an electronic device that implements the application method of the classification prediction model and the generation method of the classification prediction model according to the embodiments of the present invention. Detailed Implementation

[0053] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0054] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0055] Example 1

[0056] Figure 1 This is a flowchart illustrating an application method for a classification prediction model provided in Embodiment 1 of the present invention. This embodiment is applicable to situations where prediction is made based on collected data of the current prediction object. This method can be executed by an application device for the classification prediction model, which can be implemented in hardware and / or software and is generally configured in a computer or processor with data processing capabilities. Figure 1 As shown, the method includes:

[0057] S110. Extract features from the collected data to obtain multiple input features corresponding to the current prediction object.

[0058] The input features include categorical features and high-dimensional sparse features.

[0059] Optionally, the collected data may refer to the raw data related to the predicted object collected from the data source in order to achieve a specific classification prediction goal. For example, when it is necessary to predict the quality of a product, the collected data may include the product's production line, production time, production process, monitoring information during the production process, etc. The above is only an example.

[0060] Optionally, the collected data may include multiple features of the current prediction object, but not all of these features need to be used for classification prediction. Therefore, input features can be filtered from the collected data according to the feature types specified in advance during the model generation process, thereby retaining information in the data that is valuable for prediction and removing redundant or irrelevant information.

[0061] Optionally, the prediction object can refer to the specific individual or entity that the prediction task targets. For example, when predicting the quality of a product, the product can be used as the current prediction object.

[0062] Optionally, categorical features can refer to features that take values ​​of discrete categories, with no clear numerical relationship between their values, such as product manufacturers, product names, etc.; high-dimensional sparse features can refer to features with high dimensionality, that is, a large number of features, and most of them take values ​​of 0.

[0063] S120. Convert the discrete values ​​in the categorical features into numerical representations to generate input data for the classification prediction model.

[0064] Optionally, during the training of the classification prediction model, discrete values ​​are converted into numerical representations. Although the numerical representations of each discrete value may not be completely consistent during training, the average of all numerical representations of the same discrete value during the last training process is calculated and used as the numerical representation corresponding to that discrete value. This numerical representation is then recorded, and in practical applications, the corresponding numerical representation can be directly found based on the discrete value.

[0065] Optionally, after converting the categorical features with discrete values ​​into numerical representations, they can be combined with high-dimensional sparse features to generate the input data for the classification prediction model.

[0066] S130. Input the input data into the pre-trained classification prediction model and obtain the classification result output by the classification prediction model.

[0067] Optionally, the classification prediction model is a model built based on training data to output classification results. The classification prediction model generated in this application supports categorical features and high-dimensional sparse features, and can directly make predictions without complex calculations on the features.

[0068] Optionally, the classification result can refer to the output of the classification prediction model on the input data, or it can be a category label used to determine the probability that the current predicted object belongs to the target category.

[0069] The optimal parameter set used to generate the classification prediction model is obtained by training the initial classification model for multiple rounds; the training sample data used for training the initial classification model for multiple rounds is obtained by preprocessing and discrete value transformation of historical data of multiple objects.

[0070] Optionally, the optimal parameter set can refer to a set of parameters that optimizes model performance, determined through multiple rounds of parameter search and evaluation during model training; training sample data can refer to the dataset used for model training, which is generated from historical data of multiple objects after preprocessing and discrete value transformation, and includes input features and corresponding target values.

[0071] Optionally, the method for generating the classification prediction model may include:

[0072] Historical data of multiple objects is acquired and preprocessed to generate initial sample data; wherein, the historical data includes multiple initial features of the objects and target values, and the initial features include categorical features and high-dimensional sparse features;

[0073] The discrete values ​​in the initial sample data are converted into numerical representations to generate training sample data;

[0074] Based on the training sample data and the parameter search space, the initial classification model is trained in multiple rounds, and after each round of training, the score value corresponding to the parameter set used in the current round of training is obtained.

[0075] Once the training termination condition is met, the optimal parameter set is determined based on each score value, and the classification prediction model is trained using the optimal parameter set.

[0076] The technical solution of this invention extracts features from the collected data to obtain multiple input features corresponding to the current prediction object, converts the discrete values ​​in the categorical features into numerical representations, generates input data for the classification prediction model, and inputs the input data into the pre-trained classification prediction model to obtain the classification result output by the classification prediction model. This method enables accurate classification prediction of input features containing categorical features and high-dimensional sparse features, effectively reducing the computational load of feature processing during the prediction process, thereby improving classification efficiency.

[0077] Example 2

[0078] Figure 2This is a flowchart of a method for generating a classification prediction model according to Embodiment 2 of the present invention. This embodiment is applicable to situations where a classification prediction model is trained based on historical data of multiple objects, and the classification prediction model has the ability to predict data containing categorical features and high-dimensional sparse features. This method can be executed by a device for generating the classification prediction model, which can be implemented in hardware and / or software, and is generally configured in a computer or processor with data processing capabilities. Figure 2 As shown, the method includes:

[0079] S210. Obtain historical data of multiple objects, preprocess the historical data, and generate initial sample data.

[0080] The historical data includes multiple initial features of the object and the target value. The initial features include categorical features and high-dimensional sparse features.

[0081] Optionally, historical data for multiple objects can refer to all recorded data related to multiple objects over a past period, including multiple initial features of the objects and target values. Initial feature values ​​can refer to the collected features of the objects, and target values ​​can refer to the actual result values. For example, when a classification prediction model is used to predict the quality of a product, the object can be the product, and the initial features can include some features that can be collected for each product. During the model training phase, the type of initial features is not limited and can be all features that can be collected for the object. The target value can be the actual quality of each product. For example, the target value for a high-quality product can be recorded as 0, and the target value for a low-quality product can be recorded as 1. This is only an example and is not specifically limited.

[0082] Optionally, categorical features can refer to features that take values ​​of discrete categories, with no clear numerical relationship between their values, such as product manufacturers, product names, etc.; high-dimensional sparse features can refer to features with high dimensionality, that is, a large number of features, and most of them take values ​​of 0.

[0083] Optionally, preprocessing may include handling duplicate values ​​and handling missing values. Handling duplicate values ​​may refer to deleting completely duplicate samples when the contents of two historical data records are completely identical, so as to avoid data redundancy and cause model learning bias. Handling missing values ​​may include treating missing values ​​as independent categories for features with a low proportion of missing values, such as less than 80%, without additional filling. For features with a high proportion of missing values, such as more than 80%, the feature column may be directly deleted.

[0084] S220. Convert the discrete values ​​in the initial sample data into numerical representations to generate training sample data.

[0085] Optionally, for discrete categorical features, this application can directly process categorical data based on principled encoding of effective target statistics, converting each discrete value of the categorical feature into a meaningful numerical value that reflects the statistical correlation between the categorical value and the target variable. Compared with the one-hot encoding used in the prior art, which is prone to feature dimension explosion when encountering a large number of categorical features, this application can compress categorical features into a numerical feature, thereby accelerating training data and improving model performance, while reducing training resource overhead.

[0086] The inventors of this application considered that if the average target value of the category value on the entire training set is used to encode all samples, the model will refer to its own target value information during training, resulting in severe overfitting and poor generalization ability. Therefore, this application introduces a random permutation order to simulate time, ensuring that the encoding calculation only relies on historical data. Since the encoding calculation strictly excludes the target value of the current sample itself and relies entirely on its historical sample information, the target leakage problem can be completely solved, and the generalization ability can be significantly improved.

[0087] Converting the discrete values ​​in the initial sample data into numerical representations to generate training sample data may include:

[0088] Obtain the original index sequence of each initial sample data, and reorder the original index sequence using a random algorithm to generate a random index sequence;

[0089] Based on the random index sequence, starting from the first initial sample data, each initial sample data is traversed, and the discrete values ​​of categorical features in each initial sample data are converted into numerical representations in turn.

[0090] Optionally, the original index sequence can refer to the position number sequence of the initial sample data arranged in the original order, such as the index of samples sorted by collection time. The original index may contain an order pattern that is unrelated to the target value. Direct use will lead to feature encoding deviation. Therefore, it is necessary to break the inherent order by random rearrangement.

[0091] Optionally, a fair shuffle algorithm can be used to reorder the original index sequence. The random index sequence can be represented as φ = (φ1, φ2, ..., φi, ..., φn). The random index sequence can actually define the arrival time of each sample. For example, sample φi can be regarded as the sample that arrives at time point i, while φ1 to φi-1 can be used as its historical data.

[0092] Optionally, each sample can be processed sequentially according to the random index sequence to generate numerical codes for the discrete values ​​of its categorical features. During the traversal, the model will dynamically accumulate the historical statistics of each category value to ensure that the code reflects the relationship between the category and the target in real time, while maintaining randomness and unbiasedness.

[0093] The advantages of this setup are: by introducing a random permutation order to simulate time, it ensures that the calculation encoding relies only on historical data, avoiding overfitting caused by the model seeing its own label in advance, thus completely solving the target leakage problem and significantly improving generalization ability. By compressing the categorical features into a numerical feature, the association between the category and the target can be reflected in real time through encoding, while maintaining randomness and unbiasedness.

[0094] S230. Based on the training sample data and the parameter search space, the initial classification model is trained in multiple rounds, and after each round of training, the score value corresponding to the parameter set used in the current round of training is obtained.

[0095] Optionally, the parameter search space can refer to the set of values ​​for model parameters. It is the search boundary for parameter optimization and includes key parameters that have a significant impact on model performance and their possible value ranges. For example, the parameter search space can specify various parameter types, such as tree depth, learning rate, regularization term, sampling diversity, etc., thereby limiting the set of value ranges for each feature, such as tree depth: 4-10, learning rate: 0.01-0.3; regularization term: 1-10; sampling diversity 0-1. This is only an example.

[0096] This process involves training the initial classification model multiple times based on the training sample data and the parameter search space. After each training round, a score value corresponding to the parameter set used in the current training round is obtained, which may include:

[0097] Before each round of training begins, the current parameter group is selected in the parameter search space according to the parameter sampling strategy, and the training sample data is divided into multiple data groups.

[0098] In each set of data, one set is selected as the validation set and the remaining sets are used as the training set. Based on the current parameter set and the selected validation and training sets, multiple training sessions are conducted, and a single score value is obtained for each training session.

[0099] Calculate the average score of each individual test in the current training round as the score for the current parameter group, and update the parameter sampling strategy based on the score for the current parameter group so that the parameter group can be selected in the next training round according to the updated parameter sampling strategy.

[0100] Optionally, the parameter sampling strategy can refer to an algorithm that selects parameter combinations from the parameter search space, and the current parameter set can refer to the parameter set used for the current training round.

[0101] Optionally, the training sample data can be divided into multiple data groups on an even basis. In each round of training, one data group is selected as the validation set, and the remaining data groups are used as the training set. Training is performed once using the current parameter group. Then, the next data group is selected as the validation set, and the remaining data groups are used as the training set. Training is performed again using the current parameter group, until all data groups have been used as validation sets for training once. Then, training for that round stops. Specifically, the training samples can be divided into K mutually exclusive subsets. In each training, K-1 data groups are used as the training set and 1 data group is used as the validation set. This process is repeated K times to form one round of model training.

[0102] Optionally, during a single round of model training, each time a training session is completed based on the current validation set and training set, a single-session score is calculated. The score can be obtained by calculating the area under the receiver operating characteristic curve. Specifically, the score of each of the K training sessions can be calculated as the single-session score.

[0103] The advantage of this setup is that cross-validation can reduce the impact of randomness in sample splitting, especially in scenarios where the proportion of a single target value is low, such as the detection of non-conforming products. It avoids the extreme case where no samples are lost in the validation set due to a single split. By calculating the average score, the random fluctuations of a single validation can be offset, and the performance of the parameter group can be evaluated more objectively.

[0104] S240. When the training termination condition is met, determine the optimal parameter set based on each score value, and use the optimal parameter set to train the classification prediction model.

[0105] Optionally, the training termination condition, as the criterion for terminating multiple rounds of training, can be determined when the score changes little over several consecutive rounds or when the maximum number of iterations is reached.

[0106] Optionally, the parameter groups can be divided into high-score and low-score intervals based on their rating values. For example, the parameter groups with rating values ​​in the top 25% can be divided into high-score intervals, and the remaining parameter groups can be divided into low-score intervals. This is just an example.

[0107] Optionally, the optimal parameter set can be selected based on the distribution of training parameter values ​​in the high-score interval.

[0108] The technical solution of this invention obtains historical data of multiple objects, preprocesses the historical data to generate initial sample data, converts the discrete values ​​in the initial sample data into numerical representations to generate training sample data, trains the initial classification model multiple times based on the training sample data and the parameter search space, and obtains the score value corresponding to the parameter set used in the current training round after each round of training. When the training termination condition is met, the optimal parameter set is determined based on each score value, and the classification prediction model is trained using the optimal parameter set. This method optimizes the handling of categorical features and missing data by the classification prediction model, effectively prevents model overfitting, improves the robustness and generalization ability of the model, and enables the classification prediction model to classify and predict a large number of categorical features and high-dimensional sparse features. It reduces the impact of randomness in sample partitioning, avoids the extreme case where no samples are lost in the validation set due to a single partition, and offsets the random fluctuations of a single validation by calculating the average score, thus more objectively evaluating the performance of the parameter set and intelligently selecting the optimal parameter set, thereby improving the prediction accuracy and stability of the classification prediction model.

[0109] Example 3

[0110] Figure 3 This is a flowchart illustrating a method for generating a classification prediction model according to Embodiment 3 of the present invention. This embodiment, based on Embodiment 2, specifically describes the method for generating a classification prediction model. Figure 3 As shown, the method includes:

[0111] S310. Obtain historical data of multiple objects and perform duplicate detection on each historical data. When duplicate historical data is detected, perform deduplication operation.

[0112] Optionally, by comparing samples in historical data, completely duplicate records can be identified, such as the same features and target values ​​of the same product being stored multiple times, and redundant duplicates can be deleted.

[0113] S320. Based on the historical data of each object, detect the missing proportion of each initial feature. When the missing proportion of the target initial feature is detected to be greater than the proportion threshold, delete the target initial feature in each historical data to obtain the initial sample data of each object.

[0114] Optionally, the classification prediction model proposed in this application can be implemented based on the gradient boosting decision tree machine learning algorithm. Since the missing value can be regarded as an independent category when constructing the decision tree based on this machine learning algorithm, and its gain is calculated at each segmentation, the initial features with a certain number of missing values ​​do not need to be processed, and only the initial features with more missing values ​​need to be deleted.

[0115] The advantage of this setup is that duplicate data can cause the model to overlearn on duplicate samples, leading to prediction bias. By deleting duplicate samples, redundant samples can be removed, ensuring that the data is representative. At the same time, initial features with a high missing rate have low feature information value. Therefore, retaining only initial features with certain information value can effectively improve the training efficiency of the model and effectively reduce the collection of redundant features during application.

[0116] S330. Obtain the original index sequence of each initial sample data, and reorder the original index sequence using a random algorithm to generate a random index sequence.

[0117] S340. Based on the random index sequence, starting from the first initial sample data, traverse each initial sample data and convert the discrete values ​​of the categorical features in each initial sample data into numerical representations.

[0118] This process involves iterating through each initial sample data point based on a random index sequence, starting from the first initial sample data point, and sequentially converting the discrete values ​​of categorical features in each initial sample data point into numerical representations. This may include:

[0119] When traversing the initial target sample data, determine the target value of the target categorical feature in the initial target sample data; where the target value is a discrete value.

[0120] Based on the target value of the target category feature, determine the first statistical value and the second statistical value respectively, and calculate the encoding value of the target category feature based on the first statistical value and the second statistical value. Then, replace the target value with the calculated encoding value.

[0121] Based on the target value of the target category feature and the target value of the initial sample data, the first and second statistical values ​​are updated for use when performing discrete value transformation on the next initial sample data.

[0122] Optionally, the target initial sample data can refer to the currently traversed sample, the target categorical feature can refer to the categorical feature to be transformed in the target sample, and the target value can be a specific discrete value of the target categorical feature. For example, when the target categorical feature is a product manufacturer, the target value can be the specific name of the manufacturer.

[0123] Optionally, taking a random index sequence φ=(φ1,φ2,…,φi,…,φn) as an example, if the current traversal reaches sample φi, then the target initial sample is φi. For the target categorical feature k, if the discrete value is c, then c is encoded, and the encoded value replaces the c value in the target initial sample.

[0124] Optionally, the first statistical value can refer to the total number of times the target value appears before the initial target sample. Continuing the previous example, specifically, it can refer to the total number of times the target categorical feature k takes the value c in samples φ1 to φi-1. The second statistical value can refer to the sum of the target values ​​of all samples where the target categorical feature is the target value before the initial target sample. Specifically, it can refer to the sum of the target values ​​of all samples where the target categorical feature k takes the value c in samples φ1 to φi-1. For example, if the categorical feature k of samples φ1 to φi-1 has three initial sample data with the value c, and the target values ​​corresponding to the three initial sample data are 0, 1, and 1 respectively, then the second statistical value is 0+1+1=2. This is only an example for illustration.

[0125] Specifically, the encoded value of the target categorical feature k can be calculated using the following formula:

[0126] encoded_value = (sum[c]+a×P) / (count[c]+a);

[0127] Where encoded_value is the calculated encoded value of the target categorical feature, count[c] is the first statistical value, which is the total number of times the target value c appears before the initial target sample, sum[c] is the second statistical value, which is the sum of the target values ​​of all samples with the target value c before the initial target sample, P is the preset prior value, which can be the average of the target values ​​of all initial sample data, and a is the preset prior weight, which is a value greater than 0. The larger a is, the closer the encoded value is to the prior value P, and the smaller a is, the more the encoded value depends on historical data.

[0128] The advantage of this setup is that by calculating the encoding in the above way, we can avoid the problem of feature dimension explosion caused by one-hot encoding in existing technologies when there are a large number of categorical features. It can rely only on historical data during the encoding process, solve the target leakage problem, and thus avoid overfitting during model training and improve the model's generalization ability.

[0129] S350. Before the start of each training round, select the current parameter group in the parameter search space according to the parameter sampling strategy, and divide the training sample data into multiple data groups.

[0130] S360. Select one data group from each data group as the validation set and the remaining data groups as the training set. Perform multiple training sessions based on the current parameter group and the selected validation and training sets, and obtain the single score value for each training session.

[0131] S370. Calculate the average value of the single score in the current training round as the score value corresponding to the current parameter group, and update the parameter sampling strategy according to the score value corresponding to the current parameter group so that the parameter group can be selected according to the updated parameter sampling strategy in the next training round.

[0132] S380. When the changes in the score values ​​of multiple consecutive parameter groups are all within the target interval or the maximum number of iterations is reached, the training termination condition is determined to be met.

[0133] Optionally, the target interval is the minimum acceptable change threshold, such as 0.01. If the change in the score value is less than the target interval for a specified number of consecutive rounds, it is determined that the model performance has not improved, and training is stopped, thereby avoiding invalid iterations or overtraining and effectively reducing resource waste.

[0134] S390. Based on the magnitude of each score value, divide the parameter group into a high group and a low group, and determine the optimal value of each parameter based on the actual parameter value of each parameter in the high group and the low group, and form the optimal parameter group by combining the optimal values ​​of each parameter.

[0135] Optionally, the probability of each actual parameter value of the target parameter appearing in the high group and the low group can be counted, and the actual parameter value with the highest probability of appearing in the high group can be selected as the optimal value of the target parameter.

[0136] S3100. Using the optimal parameter set, a classification prediction model is trained.

[0137] The advantage of this setting is that by determining that the training termination condition is met when the changes in the scores of multiple consecutive parameter groups are all within the target range or the maximum number of iterations is reached, the training efficiency and performance can be balanced through the termination condition, avoiding invalid iterations or overtraining. Thus, training can be terminated even if it is not necessary to save training resources. This ensures the statistical significance of the parameters and that the parameters can be adapted to the specific prediction task. Ultimately, the trained model can capture the influence of potential patterns and features in the data while taking into account generalization ability.

[0138] Optionally, after training the classification prediction model using the optimal parameter set, it may also include:

[0139] Based on the feature types in each training sample data, determine the input feature types required for the application of the classification prediction model, and calculate the average value of each category feature as the corresponding value for the input feature in the application.

[0140] Optionally, after training the classification prediction model using the optimal parameter set, it may also include:

[0141] Based on the training status of the training sample data, the importance index corresponding to each feature type is calculated respectively; among which, the importance index includes the contribution of model structure, the dependence of model performance, and the contribution of result allocation;

[0142] Based on importance metrics, generate visualizations to provide users with the importance of each feature type.

[0143] The advantage of this setup is that users can clearly see which feature types are more likely to lead to the target classification result by viewing the visualization. For example, they can see which features are more likely to result in a product being of inferior quality, thus providing a reference for subsequent user decisions.

[0144] The technical solution of this invention obtains historical data of multiple objects, preprocesses the historical data to generate initial sample data, converts the discrete values ​​in the initial sample data into numerical representations to generate training sample data, trains the initial classification model multiple times based on the training sample data and the parameter search space, and obtains the score value corresponding to the parameter set used in the current training round after each round of training. When the training termination condition is met, the optimal parameter set is determined based on each score value, and the classification prediction model is trained using the optimal parameter set. This method optimizes the handling of categorical features and missing data by the classification prediction model, effectively prevents model overfitting, improves the robustness and generalization ability of the model, and enables the classification prediction model to classify and predict a large number of categorical features and high-dimensional sparse features. It reduces the impact of randomness in sample partitioning, avoids the extreme case where no samples are lost in the validation set due to a single partition, and offsets the random fluctuations of a single validation by calculating the average score, thus more objectively evaluating the performance of the parameter set and intelligently selecting the optimal parameter set, thereby improving the prediction accuracy and stability of the classification prediction model.

[0145] Example 4

[0146] Figure 4 This is a schematic diagram of the structure of an application device for a classification prediction model provided in Embodiment 4 of the present invention. Figure 4 As shown, the device includes: an input feature acquisition module 410, an input data generation module 420, and a classification result acquisition module 430.

[0147] The input feature acquisition module 410 is used to extract features from the collected data and obtain multiple input features corresponding to the current prediction object; wherein, the input features include categorical features and high-dimensional sparse features.

[0148] The input data generation module 420 is used to convert the discrete values ​​in the categorical features into numerical representations to generate input data for the classification prediction model.

[0149] The classification result acquisition module 430 is used to input the input data into a pre-trained classification prediction model and obtain the classification result output by the classification prediction model.

[0150] The optimal parameter set used to generate the classification prediction model is obtained by training the initial classification model for multiple rounds; the training sample data used for training the initial classification model for multiple rounds is obtained by preprocessing and discrete value transformation of historical data of multiple objects.

[0151] The technical solution of this invention extracts features from the collected data to obtain multiple input features corresponding to the current prediction object, converts the discrete values ​​in the categorical features into numerical representations, generates input data for the classification prediction model, and inputs the input data into the pre-trained classification prediction model to obtain the classification result output by the classification prediction model. This method enables accurate classification prediction of input features containing categorical features and high-dimensional sparse features, effectively reducing the computational load of feature processing during the prediction process, thereby improving classification efficiency.

[0152] The application device for the classification prediction model provided in the embodiments of the present invention can execute the application method of the classification prediction model provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0153] Example 5

[0154] Figure 5 This is a schematic diagram of the structure of a classification prediction model generation device provided in Embodiment 5 of the present invention. Figure 5 As shown, the device includes: a preprocessing module 510, a discrete value conversion module 520, a model training module 530, and an optimal parameter set selection module 540.

[0155] The preprocessing module 510 is used to acquire historical data of multiple objects and preprocess the historical data to generate initial sample data; wherein, the historical data includes multiple initial features of the objects and target values, and the initial features include categorical features and high-dimensional sparse features.

[0156] The discrete value conversion module 520 is used to convert the discrete values ​​in the initial sample data into numerical representations to generate training sample data.

[0157] The model training module 530 is used to train the initial classification model in multiple rounds based on the training sample data and the parameter search space, and to obtain the score value corresponding to the parameter set used in the current round of training after each round of training.

[0158] The optimal parameter set selection module 540 is used to determine the optimal parameter set based on each score value when the training termination condition is met, and to train the classification prediction model using the optimal parameter set.

[0159] The technical solution of this invention obtains historical data of multiple objects, preprocesses the historical data to generate initial sample data, converts the discrete values ​​in the initial sample data into numerical representations to generate training sample data, trains the initial classification model multiple times based on the training sample data and the parameter search space, and obtains the score value corresponding to the parameter set used in the current training round after each round of training. When the training termination condition is met, the optimal parameter set is determined based on each score value, and the classification prediction model is trained using the optimal parameter set. This method optimizes the handling of categorical features and missing data by the classification prediction model, effectively prevents model overfitting, improves the robustness and generalization ability of the model, and enables the classification prediction model to classify and predict a large number of categorical features and high-dimensional sparse features. It reduces the impact of randomness in sample partitioning, avoids the extreme case where no samples are lost in the validation set due to a single partition, and offsets the random fluctuations of a single validation by calculating the average score, thus more objectively evaluating the performance of the parameter set and intelligently selecting the optimal parameter set, thereby improving the prediction accuracy and stability of the classification prediction model.

[0160] Based on the above embodiments, the preprocessing module 510 can be specifically used for:

[0161] Retrieve historical data for multiple objects and perform duplicate checks on each piece of historical data. When duplicate historical data is detected, perform a deduplication operation.

[0162] Based on the historical data of each object, the missing proportion of each initial feature is detected. When the missing proportion of the target initial feature is detected to be greater than the proportion threshold, the target initial feature is deleted from each historical data to obtain the initial sample data of each object.

[0163] Based on the above embodiments, the discrete value conversion module 520 may include:

[0164] The sorting unit is used to obtain the original index sequence of each initial sample data and reorder the original index sequence using a random algorithm to generate a random index sequence;

[0165] The value conversion unit is used to traverse each initial sample data starting from the first initial sample data according to the random index sequence, and convert the discrete values ​​of the categorical features in each initial sample data into numerical representations in turn.

[0166] Based on the above embodiments, the value conversion unit can be specifically used for:

[0167] When traversing the initial target sample data, determine the target value of the target categorical feature in the initial target sample data; where the target value is a discrete value.

[0168] Based on the target value of the target category feature, determine the first statistical value and the second statistical value respectively, and calculate the encoding value of the target category feature based on the first statistical value and the second statistical value. Then, replace the target value with the calculated encoding value.

[0169] Based on the target value of the target category feature and the target value of the initial sample data, the first and second statistical values ​​are updated for use when performing discrete value transformation on the next initial sample data.

[0170] Based on the above embodiments, the model training module 530 can be specifically used for:

[0171] Before each round of training begins, the current parameter group is selected in the parameter search space according to the parameter sampling strategy, and the training sample data is divided into multiple data groups.

[0172] In each set of data, one set is selected as the validation set and the remaining sets are used as the training set. Based on the current parameter set and the selected validation and training sets, multiple training sessions are conducted, and a single score value is obtained for each training session.

[0173] Calculate the average score of each individual test in the current training round as the score for the current parameter group, and update the parameter sampling strategy based on the score for the current parameter group so that the parameter group can be selected in the next training round according to the updated parameter sampling strategy.

[0174] Based on the above embodiments, the optimal parameter set selection module 540 can be specifically used for:

[0175] When the changes in the score values ​​of multiple consecutive parameter groups are all within the target interval or the maximum number of iterations is reached, the training termination condition is determined to be met.

[0176] Based on the magnitude of each score, the parameter group is divided into a high group and a low group. Based on the actual parameter values ​​of each parameter in the high group and the low group, the optimal value of each parameter is determined, and the optimal values ​​of each parameter are combined to form the optimal parameter group.

[0177] The classification prediction model generation apparatus provided in this embodiment of the invention can execute the classification prediction model generation method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0178] Example 6

[0179] Figure 6 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0180] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0181] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0182] Processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the method for applying classification prediction models and the method for generating classification prediction models as described in embodiments of the present invention.

[0183] In some embodiments, the methods for applying and generating classification prediction models can be implemented as computer programs tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the methods for applying and generating classification prediction models described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the methods for applying and generating classification prediction models by any other suitable means (e.g., by means of firmware).

[0184] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0185] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0186] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0187] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0188] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0189] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0190] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0191] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for applying a classification prediction model, characterized in that, The method comprises: performing feature extraction on the collected data to obtain a plurality of input features corresponding to the current prediction object; wherein the input features comprise category type features and high-dimensional sparse features; converting discrete values in the category type features into numerical representations to generate input data for the classification prediction model; inputting the input data into the pre-trained classification prediction model to obtain a classification result output by the classification prediction model. The optimal parameter set for generating the classification prediction model is obtained after multiple rounds of training of an initial classification model; the training sample data for the multiple rounds of training of the initial classification model is obtained by preprocessing and converting discrete values of the historical data of the plurality of objects.

2. A method of generating a classification prediction model, the method comprising: The method comprises: obtaining historical data of a plurality of objects and preprocessing the historical data to generate initial sample data; wherein the historical data comprises a plurality of initial features of the objects and target values, and the initial features comprise category type features and high-dimensional sparse features; converting discrete values in the initial sample data into numerical representations to generate training sample data; performing multiple rounds of training of an initial classification model according to the training sample data and a parameter search space, and obtaining a score value corresponding to a parameter set used in the current round of training after each round of training ends; when it is determined that a training end condition is met, determining an optimal parameter set according to the score values, and training a classification prediction model using the optimal parameter set.

3. The method of claim 2, wherein, Obtaining historical data of a plurality of objects and preprocessing the historical data to generate initial sample data comprises: obtaining historical data of a plurality of objects and performing repeated detection on each historical data, and performing a de-duplication operation when repeated historical data is detected; detecting a missing proportion of each initial feature according to the historical data of each object, and deleting a target initial feature in each historical data when it is detected that the missing proportion of the target initial feature is greater than a proportion threshold, to obtain initial sample data of each object.

4. The method of claim 2, wherein, Converting discrete values in the initial sample data into numerical representations to generate training sample data comprises: obtaining an original index sequence of each initial sample data, and reordering the original index sequence by a random algorithm to generate a random index sequence; according to the random index sequence, traversing each initial sample data starting from a first initial sample data, and converting discrete values of category type features in each initial sample data into numerical representations in sequence.

5. The method of claim 4, wherein, According to the random index sequence, traversing each initial sample data starting from a first initial sample data, and converting discrete values of category type features in each initial sample data into numerical representations in sequence, comprises: when a target initial sample data is traversed, determining a target value of a target category type feature in the target initial sample data; wherein the target value is a discrete value; determining a first statistical value and a second statistical value according to the target value of the target category type feature, and calculating an encoding value of the target category type feature according to the first statistical value and the second statistical value, and replacing the target value with the calculated encoding value; According to the target value of the target category type feature and the target value of the target initial sample data, the first statistical value and the second statistical value are updated for use in the next initial sample data for discrete value conversion.

6. The method of claim 2, wherein, According to the training sample data and the parameter search space, the initial classification model is trained for multiple rounds, and after each round of training, a score value corresponding to the parameter group used in the current round of training is obtained, including: Before each round of training begins, according to the parameter sampling strategy, the current parameter group is selected in the parameter search space, and the training sample data is divided into multiple data groups; In each data group, one of the data groups is selected as a validation set, and the remaining data groups are used as training sets, and according to the current parameter group and each selected validation set and training set, multiple training is performed, and a single score value for each training is obtained; The average of the single score values in the current round of training is calculated as the score value corresponding to the current parameter group, and the parameter sampling strategy is updated according to the score value corresponding to the current parameter group, so that in the next round of training, the parameter group is selected according to the updated parameter sampling strategy.

7. The method of claim 2, wherein, When it is determined that the training end condition is met, the optimal parameter group is determined according to the score values, including: When the change amount of the score values of the continuous multiple parameter groups is within the target interval or reaches the maximum iteration number, it is determined that the training end condition is met; According to the size of each score value, the parameter groups are divided into high-score groups and low-score groups, and the optimal values of each parameter are determined according to the actual parameter values of each parameter in the high-score groups and the low-score groups, and the optimal values of each parameter are combined to form the optimal parameter group.

8. An electronic device, comprising: The electronic device includes: at least one processor; and a memory communicatively connected with the at least one processor; wherein The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the application method of the classification prediction model of claim 1 or the generation method of the classification prediction model of any one of claims 2-7.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions for causing the processor to execute the application method of the classification prediction model of claim 1 or the generation method of the classification prediction model of any one of claims 2-7.

10. A computer program product, characterised in that, The computer program product includes a computer program that, when executed by a processor, implements the application method of the classification prediction model according to claim 1 or the generation method of the classification prediction model according to any one of claims 2-7.

Images