Machine learning-based model training and classification method and device, equipment and medium

Abstract

The invention provides a model training and classification method and device based on machine learning, equipment and a medium, and belongs to the technical field of machine learning. The method comprises the steps of obtaining multiple pieces of sample data of a target object in a first time period; dividing each piece of sample data into a plurality of negative samples and a plurality of positive samples; obtaining multiple groups of training sets and test sets; adopting a Bayesian optimization algorithm to obtain multiple groups of hyper-parameter values, and correspondingly setting each group of hyper-parameter values as hyper-parameters of the gradient boosting decision tree; inputting each group of training sets into a corresponding target gradient boosting decision tree, and calculating a curve lower area corresponding to each target gradient boosting decision tree; and determining an object classification model from the plurality of target gradient lifting decision trees according to the area under the curve. In this way, the Bayesian optimization algorithm is adopted to obtain multiple groups of hyper-parameter values, multiple groups of training sets are combined to train multiple target gradient boosting decision trees, the object classification model is preferentially determined, the training method is simple, and the classification accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a model training and classification method, device, equipment and medium based on machine learning. Background Art

[0002] Machine learning involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It is a multi-disciplinary cross-disciplinary subject that continues to be widely used in the industrial field. For example, machine learning methods are used to predict the characteristic data of the target object. Supervised learning is often used for training. Since supervised learning often encounters the problem of unbalanced number of samples of different categories, the accuracy of predicting and classifying the characteristic data of the target object is relatively low. Summary of the invention

[0003] In order to solve the above technical problems, the embodiments of the present application provide a model training, classification method, device, equipment and medium based on machine learning.

[0004] In a first aspect, an embodiment of the present application provides a model training method based on machine learning, the method comprising: Acquire a plurality of sample data of the target object within a first period of time, each of the sample data including a preset number of key feature data of the target object; Marking each of the sample data respectively to divide them into a plurality of negative samples and a plurality of positive samples; Acquire multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; Input each group of the training sets into the corresponding target gradient boosting decision tree, and calculate the area under the curve corresponding to each target gradient boosting decision tree; An object classification model is determined from the plurality of target gradient boosting decision trees according to the area under the curve.

[0005] In one embodiment, obtaining a plurality of sample data of a target object within a first period of time includes: Acquire multi-dimensional object feature data of the target object in a first period of time, wherein the multi-dimensional object feature data includes continuous variable data, ordered categorical variable data, and unordered categorical variable data; Normalizing the continuous variable data and the ordered categorical variable data respectively to obtain normalized continuous variable data and normalized ordered categorical variable data; Performing one-hot processing on the unordered categorical variable data to obtain binary categorical variable data; A plurality of sample data are obtained from the normalized continuous variable data, the normalized ordered categorical variable data and the binary categorical variable data according to the model feature importance factor.

[0006] In one embodiment, the step of obtaining multiple sets of training sets and test sets based on the multiple negative samples and the multiple positive samples includes: Dividing the plurality of negative samples and the plurality of positive samples into a first training set and a first test set according to a preset ratio; Divide the i-th training set into the i+1-th training set and the i+1-th test set according to the preset ratio, i≥2; The positive samples in the j-th group of training sets are oversampled by a random oversampling method, so that the number of negative samples and the number of positive samples in the j-th group of training sets are the same, j≥1.

[0007] In a second aspect, the present application embodiment provides a classification method based on machine learning, including: Acquire a plurality of sample data of the target object within a first period of time, each of the sample data including a preset number of key feature data of the target object; Marking each of the sample data respectively to divide them into a plurality of negative samples and a plurality of positive samples; Acquire multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; Input each group of the training sets into the corresponding target gradient boosting decision tree, and calculate the area under the curve corresponding to each target gradient boosting decision tree; Determining an object classification model from a plurality of the target gradient boosting decision trees according to the area under the curve; Determining multi-dimensional feature data of the object to be tested in the first time period, wherein the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; The multi-dimensional feature data of the object to be tested is input into the object classification model, and the classification data of the multi-dimensional feature data of the object to be tested is obtained through the object classification model.

[0008] In one embodiment, the step of obtaining the classification data of the multi-dimensional feature data of the object to be tested by using the object classification model includes: The multi-dimensional feature data of the object to be tested is calculated by the object classification model, and the result value obtained by the calculation is 1 or 0; If the result value is 1, the multi-dimensional feature data of the object to be tested is marked as the first category; If the result value is 0, the multi-dimensional feature data of the object to be tested is marked as the second category.

[0009] In one embodiment, the classification method based on machine learning further includes: Calculate the contribution of each key feature data of the object to be tested to the result value using the SHAP algorithm; The key feature data of each of the objects to be tested are sorted according to the contribution of each of the key feature data of the objects to be tested.

[0010] In a third aspect, an embodiment of the present application provides a model training device based on machine learning, and the model training device based on machine learning includes: A first acquisition module, used to acquire a plurality of sample data of a target object within a first period of time, each of the sample data comprising a preset number of key feature data of the target object; A marking module, used to mark each sample data separately, dividing it into a plurality of negative samples and a plurality of positive samples; A second acquisition module, used for acquiring multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A setting module is used to obtain multiple groups of hyperparameter values ​​by using a Bayesian optimization algorithm, and set each group of hyperparameter values ​​as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; A calculation module, used for inputting each group of the training sets into the corresponding target gradient boosting decision tree, and calculating the area under the curve corresponding to each target gradient boosting decision tree; The first determination module is used to determine an object classification model from a plurality of the target gradient boosting decision trees according to the area under the curve.

[0011] In a fourth aspect, an embodiment of the present application provides a classification device based on machine learning, including: A first acquisition module, used to acquire a plurality of sample data of a target object within a first period of time, each of the sample data comprising a preset number of key feature data of the target object; A marking module, used to mark each sample data separately, dividing it into a plurality of negative samples and a plurality of positive samples; A second acquisition module, used for acquiring multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A setting module is used to obtain multiple groups of hyperparameter values ​​by using a Bayesian optimization algorithm, and set each group of hyperparameter values ​​as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; A calculation module, used for inputting each group of the training sets into the corresponding target gradient boosting decision tree, and calculating the area under the curve corresponding to each target gradient boosting decision tree; A first determination module, configured to determine an object classification model from a plurality of target gradient boosting decision trees according to the area under the curve; A second determination module is used to determine multi-dimensional feature data of the object to be tested in the first time period, wherein the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; The third acquisition module is used to input the multi-dimensional feature data of the object to be detected into the object classification model, and obtain classification data of the multi-dimensional feature data of the object to be detected through the object classification model.

[0012] In a fifth aspect, an embodiment of the present application provides an electronic device comprising a memory and a processor, wherein the memory is used to store a computer program, and when the computer program is run by the processor, the computer program executes the machine learning-based model training method provided in the first aspect, or executes the machine learning-based classification method provided in the second aspect.

[0013] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when running on a processor, executes the machine learning-based model training method provided in the first aspect, or executes the machine learning-based classification method provided in the second aspect.

[0014] The model training, classification method, device, equipment and medium based on machine learning provided in the present application above obtain multiple sample data of the target object in the first time period, each of which includes a preset number of key feature data of the target object; each of the sample data is marked respectively and divided into multiple negative samples and multiple positive samples; multiple groups of training sets and test sets are obtained according to the multiple negative samples and multiple positive samples; a Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; each group of the training set is input into the corresponding target gradient boosting decision tree, and the area under the curve corresponding to each target gradient boosting decision tree is calculated; an object classification model is determined from the multiple target gradient boosting decision trees according to the area under the curve; the multi-dimensional feature data of the object to be tested in the first time period is determined, and the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; the multi-dimensional feature data of the object to be tested is input into the object classification model, and the classification data of the multi-dimensional feature data of the object to be tested is obtained through the object classification model. By adopting the Bayesian optimization algorithm to obtain multiple groups of hyperparameter values, each group of hyperparameter values ​​is set as the hyperparameter of the gradient boosting decision tree to obtain the corresponding target gradient boosting decision tree, and each target gradient boosting decision tree is trained in combination with multiple groups of training sets. The object classification model is determined based on the model effect. The training method is simple, the classification accuracy of the object classification model is improved, and the classification performance of the machine learning model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present application and should not be regarded as limiting the scope of protection of the present application. In each of the drawings, similar components are numbered similarly.

[0016] Figure 1 A schematic diagram of a process of a model training method based on machine learning provided in an embodiment of the present application is shown; Figure 2 A schematic diagram of a process of a classification method based on machine learning provided in an embodiment of the present application is shown; Figure 3 A schematic diagram of the structure of a model training device based on machine learning provided in an embodiment of the present application is shown; Figure 4 A structural schematic diagram of a classification device based on machine learning provided in an embodiment of the present application is shown.

[0017] Icons: 31-model training device based on machine learning, 301-first acquisition module, 302-marking module, 303-second acquisition module, 304-setting module, 305-calculation module, 306-first determination module, 32-classification device based on machine learning, 307-second determination module, 308-third acquisition module. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0019] The components of the embodiments of the present application generally described and shown in the drawings herein may be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0020] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present application, are intended only to indicate specific features, numbers, steps, operations, elements, components, or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing items or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations of the foregoing items.

[0021] Furthermore, the terms “first”, “second”, “third”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0022] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present application belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present application.

[0023] Example 1 The embodiment of the present application provides a model training method based on machine learning, which trains a gradient boosting decision tree in combination with multiple training sets, and determines the object classification model based on the model effect. The training method is simple and improves the classification accuracy of the object classification model.

[0024] See also Figure 1 , the model training method based on machine learning includes steps S101-S106, and each step is described below.

[0025] Step S101: acquiring a plurality of sample data of a target object within a first period of time, each of the sample data comprising a preset number of key feature data of the target object.

[0026] In this embodiment, the target object is an object in a specific space, for example, the target object can be a monitored object, or for another example, the target object can be a detained object. The first period is a preset period, for example, the first period is the period before the monitored object is imprisoned. The key feature data of the target object may include the target object's age information, growth experience feature data, educational background text information, case description text information, psychological feature data, etc. The preset number of target object key feature data can be set according to actual conditions, or selected according to the importance of the key feature data. For example, the preset number can be 62, or other values, which are not limited here.

[0027] In one embodiment, step S101 includes: Acquire multi-dimensional object feature data of the target object in a first period of time, wherein the multi-dimensional object feature data includes continuous variable data, ordered categorical variable data, and unordered categorical variable data; Normalizing the continuous variable data and the ordered categorical variable data respectively to obtain normalized continuous variable data and normalized ordered categorical variable data; Performing one-hot processing on the unordered categorical variable data to obtain binary categorical variable data; A plurality of sample data are obtained from the normalized continuous variable data, the normalized ordered categorical variable data and the binary categorical variable data according to the model feature importance factor.

[0028] In this embodiment, a minimum-maximum normalization algorithm may be used, and the formula is: A=(X-min(X)) / (max(X)-min(X)), where X represents continuous variable data, max(X) represents the maximum value in the continuous variable data, min(X) represents the minimum value of the continuous variable data, and A represents the normalized value. Continuous variable data include age, amount data, etc., and ordered categorical variable data include multi-classified psychological characteristics, segmented sentences, economic conditions, etc. The minimum-maximum normalization algorithm is used to normalize the variables within the range of 0-1. In this embodiment, the one-hot process is to convert one unordered categorical variable data (i.e., multi-classified variable) into multiple 2-classified variables.

[0029] It should be noted that the multidimensional object feature data can be divided into continuous variable data, ordered categorical variable data and unordered categorical variable data; continuous variable data is continuous, such as the age, height, weight, etc. of the target object; ordered categorical variable data, although not continuous, has a clear sorting relationship, such as education level (primary school, junior high school, high school, university, etc.); unordered categorical variable data is discrete and has no clear sorting relationship, such as gender (male, female), occupation type (teachers, doctors, workers, etc.).

[0030] In this embodiment, if the age feature in the continuous variable is normalized, it is necessary to find the minimum and maximum ages in all target object data sets. For example, for "age", assuming that the minimum value is 18 years old and the maximum value is 70 years old, and assuming that there is a target object whose age is 45 years old, use the minimum-maximum normalization formula to convert each age value to the [0, 1] interval, which is approximately 0.52.

[0031] For example, assume that the target feature information is an ordered categorical variable "education level", whose categories are "primary school", "junior high school", "senior high school", and "college". First, assign them an ordinal value from low to high, such as 0, 1, 2, 3, and then normalize these values ​​to the [0, 1] interval.

[0032] In this embodiment, one-hot encoding is used for the unsorted variable class. One-hot encoding converts the value of each category into a binary vector with 1 only in the category position and 0 in the other positions. For example, the target feature information of the current target object is the gender feature (male, female). After one-hot encoding, gender_male: 1 (if male) or 0 (if not male), gender_female: 0 (if male) or 1 (if not male).

[0033] Step S102: label each sample data, and divide it into a plurality of negative samples and a plurality of positive samples.

[0034] In this embodiment, the psychological test fields in each sample data including the target object are extracted and identified. When the extracted field is identified as the first preset key semantics, the sample data is determined to be a negative sample, and when the extracted field is identified as the second preset key semantics, the sample data is determined to be a positive sample. The first preset key semantics may be "there is indeed repentance", and the second preset key semantics may be "insufficient repentance". For example, the extraction field corresponding to "there is indeed repentance" may be "admitting the facts of a specific behavior and having a strong sense of atonement". The extraction field corresponding to "insufficient repentance" may be any of the following fields: "partially admitting the facts of a specific behavior", "not admitting the facts of a specific behavior", "admitting the facts of a specific behavior, but not considering it to be a specific behavior", "admitting the specific behavior but considering that the characterization of the specific behavior is inaccurate", "admitting the specific behavior, but considering that the discretion of the specific behavior is too heavy", "admitting the specific behavior, but considering that the judgment of the specific behavior is unfair". For example, a negative sample is marked as 0 and a positive sample is marked as 1. Each target object has an identity (ID), and each target object has corresponding target object key feature data and sample labels, as shown in the following sample data schematic table 1.

[0035] Table 1: Sample data diagram.

[0036]

[0037] Step S103, obtaining multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples.

[0038] In this implementation, the plurality of negative samples and the plurality of positive samples may be divided into a plurality of training sets and test sets according to a preset ratio, the training set being used to train the gradient boosting decision tree, and the test set being used to test the trained gradient boosting decision tree. The preset ratio may be set according to actual conditions, for example, the preset ratio may be 8:2 for the training set: the test set, which is not limited here.

[0039] In one embodiment, step S103 includes: Dividing the plurality of negative samples and the plurality of positive samples into a first training set and a first test set according to a preset ratio; Divide the i-th training set into the i+1-th training set and the i+1-th test set according to the preset ratio, i≥2; The positive samples in the j-th group of training sets are oversampled by a random oversampling method, so that the number of negative samples and the number of positive samples in the j-th group of training sets are the same, j≥1.

[0040] Due to the imbalance of samples, the random oversampling method is used to oversample the positive samples in the training set so that the ratio of positive samples to negative samples in the training set reaches 1:1.

[0041] For example, there are 1000 samples, including 100 positive samples and 900 negative samples; among the 100 positive samples, multiple (such as 10 times) draw-record-replacement random sampling is performed, and 20 samples are randomly selected each time, so that 10*20=200 positive samples are oversampled in the end; these 200 positive samples are put back into the original samples, so there are a total of 1200 samples, including 300 positive samples and 900 negative samples.

[0042] Step S104, using a Bayesian optimization algorithm to obtain multiple groups of hyperparameter values, setting each group of hyperparameter values ​​as a hyperparameter of a gradient boosting decision tree, and obtaining a corresponding target gradient boosting decision tree.

[0043] It can be understood that in this embodiment, the gradient boosting decision tree (Gradient Boosting Decision Tree, GBDT) is used for model training, and logistic regression, support vector machine, random forest, and light gradient boosting machine (Light Gradient Boosting Machine, LGBM) can also be used for model training.

[0044] After comparative analysis, in this embodiment, GBDT is selected for further optimization based on the model effect. After screening features according to feature importance, Bayesian optimization with 10-fold cross-validation is used to set multiple sets of hyperparameter values, and each set of hyperparameter values ​​is set as the hyperparameter of the gradient boosting decision tree to obtain the corresponding target gradient boosting decision tree, and the hyperparameters of GBDT are adjusted with the area under the curve (AUC) as the optimization target.

[0045] It should be noted that the gradient boosting decision tree optimizes the objective function by iteratively constructing decision trees. The core idea of ​​GBDT is to minimize the loss function through gradient descent. Feature importance is usually based on the following two methods: Frequency: the number of times a feature is used for splitting in all trees. Gain: the sum of the objective function gains brought by the feature in all trees. The common loss function for binary classification GBDT is binary logarithmic loss (Log Loss): L(y, p)=−(ylog(p)+(1−y)log(1−p)), where y is the actual label and p is the predicted value.

[0046] It is further explained that for hyperparameter adjustment, the Bayesian optimization algorithm is used to select different hyperparameter values, calculate the effect of the GBDT model under the hyperparameters, and take the best set of hyperparameters; among them, the hyperparameters include learning rate, maximum depth, maximum number of features, minimum sample leaves, minimum sample splits, number of predictors, etc.

[0047] Step S105, inputting each group of the training sets into the corresponding target gradient boosting decision tree, and calculating the area under the curve corresponding to each target gradient boosting decision tree.

[0048] For example, the hyperparameters of GBDT are adjusted with AUC as the optimization target. The previous training set is split into training set and validation set again in a ratio of 8:2, and the positive samples in the split training set are randomly oversampled to ensure the balance between positive and negative samples. Each group of training sets obtained by splitting is input into the corresponding target gradient boosting decision tree, and the area under the curve corresponding to each target gradient boosting decision tree is calculated.

[0049] It should be noted that if all calculated AUC values ​​are less than the preset AUC threshold, the model's feature_importance function is used to calculate the importance of each feature. For example, for the GBDT model, the importance of each feature is determined by calculating the label impurity that is reduced when splitting the node. According to the order of feature importance, the least important feature is deleted from the pre-trained data set in turn, and the model is retrained using the split training sets. The performance of the new model is evaluated to obtain a new AUC value. If the new AUC value is still less than the preset evaluation index threshold, repeat the above steps and continue to delete the least important feature. If the new AUC value reaches or exceeds the preset threshold, the iteration is stopped.

[0050] Step S106, determining an object classification model from the plurality of target gradient boosting decision trees according to the area under the curve.

[0051] It should be noted that the area under the curve is an important indicator for evaluating the performance of binary classification models in machine learning and data analysis. It is usually used together with the ROC curve (Receiver Operating Characteristic Curve). The range of the AUC value is [0, 1], and the larger the value, the better the classification performance of the model. Therefore, the target gradient boosting decision tree corresponding to the maximum area under the curve is selected as the object classification model to improve the binary classification accuracy of the object classification model. In addition, in this embodiment, the positive samples in the training set are randomly oversampled to ensure the balance between positive and negative samples, and an independent validation set is used to evaluate the performance of the model to avoid overfitting.

[0052] The machine learning-based model training method provided in the embodiment of the present application obtains multiple sample data of the target object within a first time period, each of the sample data includes a preset number of key feature data of the target object; each of the sample data is marked and divided into multiple negative samples and multiple positive samples; multiple groups of training sets and test sets are obtained according to the multiple negative samples and the multiple positive samples; a Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of the hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; each group of the training sets is input into the corresponding target gradient boosting decision tree, and the area under the curve corresponding to each target gradient boosting decision tree is calculated; an object classification model is determined from the multiple target gradient boosting decision trees according to the area under the curve, a Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and multiple target gradient boosting decision trees are trained in combination with multiple training sets, and the object classification model is determined based on the model effect. The training method is simple and the classification accuracy of the object classification model is improved.

[0053] Example 2 The embodiment of the present application provides a classification method based on machine learning, which trains a gradient boosting decision tree in combination with multiple groups of training sets, and determines the object classification model based on the model effect. The training method is simple, improves the classification accuracy of the object classification model, and improves the classification performance of the machine learning model.

[0054] See also Figure 2 The model training method based on machine learning includes steps S201-S208, and each step is described below.

[0055] Step S201: acquiring a plurality of sample data of a target object within a first period of time, each of the sample data comprising a preset number of key feature data of the target object.

[0056] Step S201 in this embodiment is the same as step S101 in the above-mentioned embodiment 1, and will not be described here in detail to avoid repetition.

[0057] Step S202: Mark each sample data separately to divide it into a plurality of negative samples and a plurality of positive samples.

[0058] Step S202 in this embodiment is the same as step S102 in the above-mentioned embodiment 1, and will not be described here in detail to avoid repetition.

[0059] Step S203, obtaining multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples.

[0060] Step S203 in this embodiment is the same as step S103 in the above-mentioned embodiment 1, and will not be described here in detail to avoid repetition.

[0061] Step S204, using the Bayesian optimization algorithm to obtain multiple groups of hyperparameter values, setting each group of hyperparameter values ​​as the hyperparameters of the gradient boosting decision tree, and obtaining the corresponding target gradient boosting decision tree.

[0062] Step S204 in this embodiment is the same as step S104 in the above-mentioned embodiment 1, and will not be described here in detail to avoid repetition.

[0063] Step S205: input each group of the training sets into the corresponding target gradient boosting decision tree, and calculate the area under the curve corresponding to each target gradient boosting decision tree.

[0064] Step S205 in this embodiment is the same as step S105 in the above-mentioned embodiment 1, and will not be described here in detail to avoid repetition.

[0065] Step S206, determining an object classification model from the plurality of target gradient boosting decision trees according to the area under the curve.

[0066] Step S206 in this embodiment is the same as step S106 in the above-mentioned embodiment 1, and will not be described here in detail to avoid repetition.

[0067] Step S207 , determining multi-dimensional feature data of the object to be tested in the first time period, wherein the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested.

[0068] In this embodiment, the object to be tested may be a candidate object of a specific space, for example, the object to be tested may be a candidate monitoring object, for example, the object to be tested may be a candidate detained object, etc. The key feature data of the object to be tested may include the age information, educational background text information, case description text information, psychological test data, etc. of the object to be tested. The preset number of key feature data of the object to be tested can be set according to the actual situation, or selected according to the importance of the key feature data of the object to be tested. For example, the preset number can be 80, or other values, which are not limited here.

[0069] Step S208: input the multi-dimensional feature data of the object to be tested into the object classification model, and obtain classification data of the multi-dimensional feature data of the object to be tested through the object classification model.

[0070] In this embodiment, the object classification model is a binary classification model, and the object classification model can be used to perform classification calculations on the multi-dimensional feature data of the object to be tested to obtain a binary classification result.

[0071] In one embodiment, step S208 includes: The multi-dimensional feature data of the object to be tested is calculated by the object classification model, and the result value obtained by the calculation is 1 or 0; If the result value is 1, the multi-dimensional feature data of the object to be tested is marked as the first category; If the result value is 0, the multi-dimensional feature data of the object to be tested is marked as the second category.

[0072] For example, the first category is high repentance and the second category is low repentance.

[0073] In one embodiment, the model training method based on machine learning further includes: Calculate the contribution of each key feature data of the object to be tested to the result value using the SHAP algorithm; The key feature data of each of the objects to be tested are sorted according to the contribution of each of the key feature data of the objects to be tested.

[0074] It should be noted that the feature contribution mining is performed according to the Shapley Additive exPlanations (SHAP) algorithm. The SHAP algorithm is used to calculate the contribution of the key feature data of the object to be tested to the binary classification results. The SHAP value is based on the concept of Shapley value in cooperative game theory. It can fairly distribute the differences in the binary classification results to the key feature data of each object to be tested. The key feature data of each object to be tested is sorted according to the contribution index to obtain a feature contribution ranking sequence. According to the feature contribution ranking sequence, the key feature data of the target object corresponding to a historical feature information is deleted from various training sets in sequence. After each feature is deleted, a new training set is obtained. Based on the new training set, each target gradient boosting decision tree is trained and evaluated to observe the changes in model performance.

[0075] Specifically, the Shapley value is derived from game theory and is used to measure the contribution of each target object's key feature data to the binary classification result (model prediction). SHAP provides an additive interpretation method that decomposes the model's prediction results into the sum of the contributions of each feature.

[0076] Assuming that the object classification model has M features and the predicted value of the object classification model is f(x), the decomposition formula of the SHAP value is:

[0077] in, It is the baseline value of the object classification model, usually the average prediction value of the training set.

[0078] SHAP calculates the contribution of each feature by following these steps: Step 1: Decomposition of model predictions, including: 1.1 Decompose the prediction results of the object classification model into the contribution of each feature; 1.2 For sample x, the model's predicted value f(x) is decomposed into the benchmark value And the SHAP value of each feature .

[0079] Step 2: Calculation of SHAP value, including: 2.1 Calculation of SHAP value considers all possible feature combinations and measures the contribution of each feature in different combinations; 2.2 For a feature i, its SHAP value The calculation formula is:

[0080] in, Refers to the possible combinations of all participants, here refers to the possible combinations of all features, S is the set of features, is the set containing feature i The revenue (i.e., the predicted value) is the marginal contribution of feature i to the set S, and M is the number of features of the object classification model.

[0081] In one embodiment, the model training method based on machine learning further includes: Acquire behavior description text data of the object to be tested in a second time period, and determine first classification data of the object to be tested according to the behavior description text data; Displaying a plurality of test sample questions in a preset answer database, and receiving response data of the subject to be tested to each of the test sample questions; generating second classification data of the object to be tested according to the response data; The final classification data of the object to be tested is obtained according to the classification data of the multi-dimensional feature data of the object to be tested, the first classification data and the second classification data.

[0082] In this embodiment, the second time period may be the time period after the monitored object is imprisoned, and the behavior description text data within the second time period may be text data describing the daily behavior of the object to be tested. The description text data may be annotated with corresponding scoring terms, and the corresponding test score may be calculated based on the scoring terms.

[0083] In one embodiment, determining the first classification data of the object to be tested according to the behavior description text data includes: Identify the keywords of the behavior text data, and retrieve the target behavior terms matching the keywords from a preset behavior description database; the preset behavior description database includes a mapping relationship between the behavior description text data and the behavior terms; Retrieving behavior assessment scoring terms according to the target behavior terms; The first classification data is generated according to the behavior assessment scoring terms.

[0084] Exemplarily, a first test score is obtained according to the behavioral assessment scoring terms, all first test scores are accumulated, and first classification data is determined according to the relationship between the total test score and the first preset score. If the total test score is greater than the first preset score, the first classification data indicates a high degree of repentance; if the total test score is less than the first preset score, the first classification data indicates a low degree of repentance.

[0085] In one embodiment, the model training method based on machine learning further includes: A construction module is used to construct the preset answer database, the answer database includes multiple test subsets, each test subset includes multiple test sample questions, and each test sample question has a different degree of difficulty.

[0086] In this embodiment, the test subset includes 4 test subsets. The first test subset is used to test the target object's direct understanding of specific behaviors, including test sample questions such as specific behavior facts, judgment opinions, specific behavior causes, consequences and coping methods. The second test subset is used to explore the target object or the subject to be tested's understanding and attitude towards the victim, including test sample questions such as identity recognition, impact recognition and apology expression. The third test subset is used to set different test sample questions according to the remaining sentence of the target object or the subject to be tested, focusing on their understanding and planning for themselves and the future. The fourth test subset is used to evaluate the target object or the subject to be tested's understanding of the impact of specific behaviors, including test sample questions for multiple levels such as individuals, families, and society.

[0087] In one embodiment, displaying a plurality of test sample questions in a preset answer database and receiving response data of the subject to be tested for each of the test sample questions includes: Displaying a plurality of initial test sample questions in a preset answer database, receiving first response data of the subject to be tested for each of the initial test sample questions; displaying a plurality of revised test sample questions according to the first response data, and receiving second response data of the subject to be tested for each of the revised test sample questions; the difficulty of the initial test sample questions is less than that of the revised test sample questions, and the number of the initial test sample questions is less than that of the revised test sample questions; Second classification data of the object to be detected is generated according to the first response data and the second response data.

[0088] Exemplarily, the display screen displays the test sample questions. If the assessee's answer to the initial sample questions is poor, the questions on the display screen will be adjusted to fewer and simpler revised test sample questions. If the assessee's answer to the initial sample questions is good, the revised sample questions will be more and more comprehensive.

[0089] Assign different scoring weights to each test sample question, calculate the second test score for the test sample question and the corresponding scoring weight, calculate the total score of all second test scores, and determine the second classification data based on the relationship between the total score and the second preset score. If the total score is greater than the second preset score, the second classification data indicates a high degree of repentance; if the total score is less than the second preset score, the second classification data indicates a low degree of repentance.

[0090] In this embodiment, the final classification data of the object to be tested is obtained by combining the classification data of the multi-dimensional feature data of the object to be tested, the first classification data and the second classification data, specifically including: if the classification data of the multi-dimensional feature data of the object to be tested is a high degree of repentance, the first classification data and / or the second classification data are a high degree of repentance, then the final classification data is a high degree of repentance. If the classification data of the multi-dimensional feature data of the object to be tested is a low degree of repentance, the first classification data and / or the second classification data are a low degree of repentance, then the final classification data is a low degree of repentance.

[0091] The machine learning-based classification method provided in the present embodiment obtains multiple sample data of a target object in a first time period, each of the sample data includes a preset number of key feature data of the target object; each of the sample data is marked and divided into multiple negative samples and multiple positive samples; multiple groups of training sets and test sets are obtained according to the multiple negative samples and multiple positive samples; a Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of the hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; each group of the training sets is input into the corresponding target gradient boosting decision tree, and the area under the curve corresponding to each target gradient boosting decision tree is calculated; an object classification model is determined from the multiple target gradient boosting decision trees according to the area under the curve; multi-dimensional feature data of the object to be tested in the first time period is determined, and the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; the multi-dimensional feature data of the object to be tested is input into the object classification model, and classification data of the multi-dimensional feature data of the object to be tested is obtained through the object classification model. By adopting the Bayesian optimization algorithm to obtain multiple groups of hyperparameter values, each group of hyperparameter values ​​is set as the hyperparameter of the gradient boosting decision tree to obtain the corresponding target gradient boosting decision tree, and each target gradient boosting decision tree is trained in combination with multiple groups of training sets. The object classification model is determined based on the model effect. The training method is simple, the classification accuracy of the object classification model is improved, and the classification performance of the machine learning model is improved.

[0092] Example 3 In addition, an embodiment of the present application provides a model training device based on machine learning.

[0093] like Figure 3 As shown, the model training device 31 based on machine learning includes: A first acquisition module 301 is used to acquire a plurality of sample data of a target object within a first period of time, each of the sample data includes a preset number of key feature data of the target object; A marking module 302 is used to mark each sample data, and divide it into a plurality of negative samples and a plurality of positive samples; A second acquisition module 303, used to acquire multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A setting module 304 is used to obtain multiple groups of hyperparameter values ​​using a Bayesian optimization algorithm, and set each group of hyperparameter values ​​as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; A calculation module 305 is used to input each group of the training sets into the corresponding target gradient boosting decision tree, and calculate the area under the curve corresponding to each target gradient boosting decision tree; The first determination module 306 is used to determine an object classification model from the plurality of target gradient boosting decision trees according to the area under the curve.

[0094] In one embodiment, the first acquisition module 301 is further used to acquire multi-dimensional object feature data of the target object in the first time period, wherein the multi-dimensional object feature data includes continuous variable data, ordered categorical variable data, and unordered categorical variable data; Normalizing the continuous variable data and the ordered categorical variable data respectively to obtain normalized continuous variable data and normalized ordered categorical variable data; Performing one-hot processing on the unordered categorical variable data to obtain binary categorical variable data; A plurality of sample data are obtained from the normalized continuous variable data, the normalized ordered categorical variable data and the binary categorical variable data according to the model feature importance factor.

[0095] A second acquisition module 303, used to divide the plurality of negative samples and the plurality of positive samples into a first training set and a first test set according to a preset ratio; Divide the i-th training set into the i+1-th training set and the i+1-th test set according to the preset ratio, i≥2; The positive samples in the j-th group of training sets are oversampled by a random oversampling method, so that the number of negative samples and the number of positive samples in the j-th group of training sets are the same, j≥1.

[0096] The machine learning-based model training device 31 provided in this embodiment can implement the functions corresponding to the machine learning-based model training method provided in Example 1. To avoid repetition, it will not be described here.

[0097] The machine learning-based model training device provided in the present embodiment obtains multiple sample data of the target object within a first time period, each of the sample data includes a preset number of key feature data of the target object; each of the sample data is marked and divided into multiple negative samples and multiple positive samples; multiple groups of training sets and test sets are obtained according to the multiple negative samples and the multiple positive samples; a Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of the hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; each group of the training sets is input into the corresponding target gradient boosting decision tree, and the area under the curve corresponding to each target gradient boosting decision tree is calculated; an object classification model is determined from the multiple target gradient boosting decision trees according to the area under the curve, a Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and multiple target gradient boosting decision trees are trained in combination with multiple training sets, and an object classification model is determined based on the model effect. The training method is simple and the classification accuracy of the object classification model is improved.

[0098] Example 4 In addition, an embodiment of the present application provides a classification device based on machine learning.

[0099] See also Figure 4 , the classification device 32 based on machine learning includes: A first acquisition module 301 is used to acquire a plurality of sample data of a target object within a first period of time, each of the sample data includes a preset number of key feature data of the target object; A marking module 302 is used to mark each sample data, and divide it into a plurality of negative samples and a plurality of positive samples; A second acquisition module 303, used to acquire multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A setting module 304 is used to obtain multiple groups of hyperparameter values ​​using a Bayesian optimization algorithm, and set each group of hyperparameter values ​​as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; A calculation module 305 is used to input each group of the training sets into the corresponding target gradient boosting decision tree, and calculate the area under the curve corresponding to each target gradient boosting decision tree; A first determination module 306, configured to determine an object classification model from a plurality of the target gradient boosting decision trees according to the area under the curve; A second determination module 307 is used to determine the multi-dimensional feature data of the object to be tested in the first time period, wherein the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; The third acquisition module 308 is used to input the multi-dimensional feature data of the object to be tested into the object classification model, and obtain classification data of the multi-dimensional feature data of the object to be tested through the object classification model.

[0100] In one embodiment, the third acquisition module 308 is further used to calculate the multi-dimensional feature data of the object to be tested by using the object classification model, and the calculated result value is 1 or 0; If the result value is 1, the multi-dimensional feature data of the object to be tested is marked as the first category; If the result value is 0, the multi-dimensional feature data of the object to be tested is marked as the second category.

[0101] In one embodiment, the machine learning based classification device 32 further includes: A processing module, used for calculating the contribution of each key feature data of the object to be tested to the result value by using the SHAP algorithm; The key feature data of each of the objects to be tested are sorted according to the contribution of each of the key feature data of the objects to be tested.

[0102] The machine learning-based classification device provided in the present embodiment obtains multiple sample data of a target object in a first time period, each of the sample data includes a preset number of key feature data of the target object; each of the sample data is marked and divided into multiple negative samples and multiple positive samples; multiple groups of training sets and test sets are obtained according to the multiple negative samples and multiple positive samples; a Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of the hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; each group of the training sets is input into the corresponding target gradient boosting decision tree, and the area under the curve corresponding to each target gradient boosting decision tree is calculated; an object classification model is determined from the multiple target gradient boosting decision trees according to the area under the curve; multi-dimensional feature data of the object to be tested in the first time period is determined, and the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; the multi-dimensional feature data of the object to be tested is input into the object classification model, and classification data of the multi-dimensional feature data of the object to be tested is obtained through the object classification model. By adopting the Bayesian optimization algorithm to obtain multiple groups of hyperparameter values, each group of hyperparameter values ​​is set as the hyperparameter of the gradient boosting decision tree to obtain the corresponding target gradient boosting decision tree, and each target gradient boosting decision tree is trained in combination with multiple groups of training sets. The object classification model is determined based on the model effect. The training method is simple, the classification accuracy of the object classification model is improved, and the classification performance of the machine learning model is improved.

[0103] Example 5 In addition, an embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program runs on the processor, it executes the machine learning-based model training method provided in Example 1, or executes the machine learning-based classification method provided in Example 2.

[0104] The electronic device provided in the embodiment of the present invention can execute the model training method based on machine learning provided in the above-mentioned method embodiment 1, or execute the steps of the classification method based on machine learning provided in embodiment 2. To avoid repetition, they will not be repeated here.

[0105] Example 6 The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the model training method based on machine learning provided in Example 1 is implemented, or the classification method based on machine learning provided in Example 2 is executed.

[0106] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0107] The computer-readable storage medium provided in this embodiment can implement the model training method based on machine learning provided in Example 1, or execute the classification method based on machine learning provided in Example 2. To avoid repetition, it will not be described here.

[0108] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.

[0109] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0110] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A model training method based on machine learning, characterized in that: include: Acquire a plurality of sample data of the target object within a first period of time, each of the sample data including a preset number of key feature data of the target object; Marking each of the sample data respectively to divide them into a plurality of negative samples and a plurality of positive samples; Acquire multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; Input each group of the training sets into the corresponding target gradient boosting decision tree, and calculate the area under the curve corresponding to each target gradient boosting decision tree; An object classification model is determined from the plurality of target gradient boosting decision trees according to the area under the curve.

2. The method according to claim 1, characterized in that The step of obtaining a plurality of sample data of the target object within a first period of time includes: Acquire multi-dimensional object feature data of the target object in a first period of time, wherein the multi-dimensional object feature data includes continuous variable data, ordered categorical variable data, and unordered categorical variable data; Normalizing the continuous variable data and the ordered categorical variable data respectively to obtain normalized continuous variable data and normalized ordered categorical variable data; Performing one-hot processing on the unordered categorical variable data to obtain binary categorical variable data; A plurality of sample data are obtained from the normalized continuous variable data, the normalized ordered categorical variable data and the binary categorical variable data according to the model feature importance factor.

3. The method according to claim 1, characterized in that The step of obtaining multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples includes: Dividing the plurality of negative samples and the plurality of positive samples into a first training set and a first test set according to a preset ratio; Divide the i-th training set into the i+1-th training set and the i+1-th test set according to the preset ratio, i≥2; The positive samples in the j-th group of training sets are oversampled by a random oversampling method, so that the number of negative samples and the number of positive samples in the j-th group of training sets are the same, j≥1.

4. A classification method based on machine learning, characterized in that: include: Acquire a plurality of sample data of the target object within a first period of time, each of the sample data including a preset number of key feature data of the target object; Marking each of the sample data respectively to divide them into a plurality of negative samples and a plurality of positive samples; Acquire multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A Bayesian optimization algorithm is used to obtain multiple groups of hyperparameter values, and each group of hyperparameter values ​​is set as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; Input each group of the training sets into the corresponding target gradient boosting decision tree, and calculate the area under the curve corresponding to each target gradient boosting decision tree; Determining an object classification model from a plurality of the target gradient boosting decision trees according to the area under the curve; Determining multi-dimensional feature data of the object to be tested in the first time period, wherein the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; The multi-dimensional feature data of the object to be tested is input into the object classification model, and the classification data of the multi-dimensional feature data of the object to be tested is obtained through the object classification model.

5. The method according to claim 4, characterized in that The step of obtaining the classification data of the multi-dimensional feature data of the object to be tested by using the object classification model includes: The multi-dimensional feature data of the object to be tested is calculated by the object classification model, and the result value obtained by the calculation is 1 or 0; If the result value is 1, the multi-dimensional feature data of the object to be tested is marked as the first category; If the result value is 0, the multi-dimensional feature data of the object to be tested is marked as the second category.

6. The method according to claim 5, characterized in that Also includes: Calculate the contribution of each key feature data of the object to be tested to the result value using the SHAP algorithm; The key feature data of each of the objects to be tested are sorted according to the contribution of each of the key feature data of the objects to be tested.

7. A model training device based on machine learning, characterized in that: include: A first acquisition module, used to acquire a plurality of sample data of a target object within a first period of time, each of the sample data comprising a preset number of key feature data of the target object; A marking module, used to mark each sample data separately, dividing it into a plurality of negative samples and a plurality of positive samples; A second acquisition module, used for acquiring multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A setting module is used to obtain multiple groups of hyperparameter values ​​by using a Bayesian optimization algorithm, and set each group of hyperparameter values ​​as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; A calculation module, used for inputting each group of the training sets into the corresponding target gradient boosting decision tree, and calculating the area under the curve corresponding to each target gradient boosting decision tree; The first determination module is used to determine an object classification model from a plurality of the target gradient boosting decision trees according to the area under the curve.

8. A classification device based on machine learning, characterized in that: include: A first acquisition module, used to acquire a plurality of sample data of a target object within a first period of time, each of the sample data comprising a preset number of key feature data of the target object; A marking module, used to mark each sample data separately, dividing it into a plurality of negative samples and a plurality of positive samples; A second acquisition module, used for acquiring multiple sets of training sets and test sets according to the multiple negative samples and the multiple positive samples; A setting module is used to obtain multiple groups of hyperparameter values ​​by using a Bayesian optimization algorithm, and set each group of hyperparameter values ​​as a hyperparameter of a gradient boosting decision tree to obtain a corresponding target gradient boosting decision tree; A calculation module, used for inputting each group of the training sets into the corresponding target gradient boosting decision tree, and calculating the area under the curve corresponding to each target gradient boosting decision tree; A first determination module, configured to determine an object classification model from a plurality of target gradient boosting decision trees according to the area under the curve; A second determination module is used to determine multi-dimensional feature data of the object to be tested in the first time period, wherein the multi-dimensional feature data of the object to be tested includes the preset number of key feature data of the object to be tested; The third acquisition module is used to input the multi-dimensional feature data of the object to be detected into the object classification model, and obtain classification data of the multi-dimensional feature data of the object to be detected through the object classification model.

9. An electronic device, characterized in that: It includes a memory and a processor, the memory stores a computer program, and when the computer program is run on the processor, it executes the model training method based on machine learning described in any one of claims 1-3, or executes the classification method based on machine learning described in any one of claims 4-6.

10. A computer-readable storage medium, characterized in that: It stores a computer program, which, when running on a processor, executes the model training method based on machine learning described in any one of claims 1-3, or executes the classification method based on machine learning described in any one of claims 4-6.

Images