Integrated learning performance prediction method, device, electronic device and storage medium

Abstract

The present invention discloses an integrated learning performance prediction method, device, electronic device and storage medium, the method comprising: obtaining learning behavior data of target students; using a pre-trained XGBoost model to predict the iteration information of the learning behavior data during the XGBoost model training process; performing unique hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data; inputting the binary feature vector into a pre-trained soft voting classifier to obtain the target student's performance prediction result. The present invention fuses the XGBoost model, the LR model and the RF model in the second-layer prediction, performs weighted averaging on the prediction data of the three models to obtain a fusion prediction result, solves the problem of poor prediction performance caused by the single bottom-layer model of the traditional prediction method, and effectively improves the prediction performance and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of performance prediction, and in particular to an integrated learning performance prediction method, device, electronic equipment and storage medium. Background Art

[0002] At present, with the maturity and popularity of online education platforms, a hybrid teaching model combining offline and online has emerged, which not only enriches educational theories and methods, but also fills the gap in the application of digital technology. It is very necessary to predict grades in this teaching model. By analyzing the comprehensive data of students' online activity records and offline performance this week, it is predicted whether students are likely to fail the final exam, so as to carry out early intervention. This data-driven method not only improves teachers' comprehensive understanding of students' learning status, but also effectively improves students' self-consciousness in learning, reduces the risk of failing the exam, and thus optimizes teaching results.

[0003] There have been many studies on machine learning performance prediction. After comprehensive consideration, there are still some problems that need to be solved:

[0004] 1. Generally, the performance prediction is based only on the behavioral performance data of online or offline learning, which has certain limitations when dealing with more complex and changeable data generated in the hybrid teaching model. The performance prediction model for the hybrid teaching model needs to be improved in terms of prediction performance;

[0005] 2. The acquisition dimension of student learning behavior data is relatively long, usually the data of the first half of the semester or longer, which makes it difficult to achieve timely performance prediction, resulting in insufficient time for teachers and students to adapt to the change;

[0006] 3. Generally, a single model or a stacking model is used to predict grades. Under this structure, the single meta-learner leads to poor performance of the prediction model. Summary of the invention

[0007] The main technical problem solved by the present invention is to provide an integrated learning performance prediction method, device, electronic device and storage medium to solve the technical problems that the performance prediction method in the prior art has limitations when processing more complex and changeable data generated in the hybrid teaching mode, the prediction performance needs to be improved, the data acquisition dimension is long, it is difficult to achieve timely performance prediction, and the single meta-learner under the stacking structure leads to poor performance of the prediction model.

[0008] In order to solve the above technical problems, the first aspect of the present invention provides an integrated learning performance prediction method, comprising:

[0009] Acquire learning behavior data of target students, wherein the learning behavior data includes online learning behavior characteristics and / or offline learning behavior characteristics within a certain period;

[0010] Using a pre-trained XGBoost model to predict iteration information of the learning behavior data during the XGBoost model training process, the iteration information includes leaf nodes that the learning behavior data may fall into in all decision trees iteratively generated during the XGBoost model training process;

[0011] One-hot encoding the iteration information to obtain a binary feature vector of the learning behavior data;

[0012] The binary feature vector is input into a pre-trained soft voting classifier to obtain the target student's score prediction result, wherein the soft voting classifier includes the XGBoost model, LR model and RF model in parallel to perform weighted averaging on the prediction data of the XGBoost model, LR model and RF model.

[0013] In one or more embodiments, the learning behavior characteristics include period superposition characteristics and period independence characteristics;

[0014] The step of obtaining the learning behavior data of the target student includes:

[0015] Obtaining the period-independent features of the target student in the current period, and the superposition value of the period-superposition features in the current period and the previous period, and obtaining the period-superposition features of the target student in the current period based on the superposition value of the period-superposition features of the target student in the current period and the previous period;

[0016] Determine whether the target student's period superposition characteristics and period independent characteristics in the current period have negative values ​​or are missing;

[0017] If there are negative values, re-extract the corresponding student data;

[0018] If there is any missing, delete the target student’s period superimposed features and period independent features in the current period.

[0019] By adopting the above scheme, for the students' periodic superposition features, since they gradually increase with the increase of students' learning cycles, such as the number of completed courses, course video progress, etc., the superposition value of the current cycle can be obtained by subtracting the superposition value of the previous cycle; and for the periodic independent features, since they are the features completed by students in the current cycle, such as homework scores and in-class test scores, they can be directly obtained.

[0020] In order to obtain periodic superposition features, some features may have negative values ​​when the data of adjacent periods are subtracted. At the same time, there may be missing data when the data is acquired, so data preprocessing is required to ensure data accuracy.

[0021] Based on the above scheme, features that can accurately describe students' learning behaviors in the current cycle can be obtained, which helps to improve the accuracy of subsequent performance predictions.

[0022] In one or more embodiments, the training method of the XGBoost model includes:

[0023] Acquire a sample target set and divide it into a sample training set and a sample test set, wherein the sample target set includes the learning behavior data with labeled sample labels;

[0024] Based on the grid search method, the XGBoost model is cross-validated with the sample training set to evaluate the performance of each set of parameters, and then the optimal model parameters are obtained based on the sample test set.

[0025] In one or more embodiments, the training method of the soft voting classifier includes:

[0026] Recording the first training iteration information of the sample training set in the XGBoost model training process, wherein the first training iteration information includes all decision trees iteratively generated in the XGBoost model training process and the leaf nodes into which each learning behavior data in the sample training set falls in each decision tree;

[0027] Use the trained XGBoost model to predict the second training iteration information of the sample test set in the XGBoost model training process, wherein the second training iteration information includes all decision trees iteratively generated in the XGBoost model training process and the leaf nodes that each learning behavior data in the sample test set may fall into in each decision tree;

[0028] Performing one-hot encoding on the first training iteration information and the second training iteration information to obtain a binary feature vector of each of the learning behavior data;

[0029] Based on the binary feature vectors of the learning behavior data and the labeled sample labels thereof, a binary training set and a binary test set are obtained;

[0030] The soft voting classifier is trained and tested using the binary training set and the binary test set.

[0031] In one or more embodiments, the step of training and testing the soft voting classifier using the binary training set and the binary test set includes:

[0032] Based on the grid search method, the soft voting classifier is cross-validated with the binary training set to evaluate the performance of each set of parameters, and then the optimal model parameters of the three models under the soft voting classifier are obtained based on the binary test set.

[0033] In one or more embodiments, the sample label includes 0 and 1, where 0 represents an unsatisfactory final exam score and 1 represents a satisfactory final exam score.

[0034] In one or more embodiments, the learning behavior data include online learning behavior characteristics and offline learning behavior characteristics within a certain period, the online learning behavior data include the number of tasks completed, course video progress, video viewing time, number of chapter studies, chapter test scores, and course points, and the offline learning behavior characteristics include the number of discussions participated in, discussion scores, homework scores, sign-in rates, student interaction scores, group task scores, number of votes, number of selected people, and in-class quiz scores.

[0035] In one or more embodiments, the step of performing one-hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data includes:

[0036] Arrange all decision trees iteratively generated during the XGBoost model training process in parallel according to the generation order;

[0037] The leaf node where the learning behavior data falls in the decision tree is set to 1, and the other leaf nodes are set to 0, to obtain a binary feature vector of the learning behavior data.

[0038] In order to solve the above technical problems, the second aspect of the present invention provides an integrated learning performance prediction device, comprising:

[0039] A data acquisition module, used to acquire the learning behavior data of the target students, wherein the learning behavior data includes online learning behavior characteristics and / or offline learning behavior characteristics within a certain period;

[0040] An iteration information acquisition module, used to use a pre-trained XGBoost model to predict the iteration information of the learning behavior data during the training process of the XGBoost model, wherein the iteration information includes the leaf nodes that the learning behavior data may fall into in all decision trees iteratively generated during the training process of the XGBoost model;

[0041] An encoding module, used for performing one-hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data;

[0042] A fusion prediction module, used for inputting the binary feature vector into a pre-trained soft voting classifier to obtain the score prediction result of the target student, wherein the soft voting classifier includes the XGBoost model, the LR model and the RF model in parallel to perform weighted averaging on the prediction data of the XGBoost model, the LR model and the RF model;

[0043] Wherein, the learning behavior characteristics include period superposition characteristics and period independence characteristics;

[0044] The step of obtaining the learning behavior data of the target student includes:

[0045] Obtaining the period-independent features of the target student in the current period, and the superposition value of the period-superposition features in the current period and the previous period, and obtaining the period-superposition features of the target student in the current period based on the superposition value of the period-superposition features of the target student in the current period and the previous period;

[0046] Determine whether the target student's period superposition characteristics and period independent characteristics in the current period have negative values ​​or are missing;

[0047] If there are negative values, re-extract the corresponding student data;

[0048] If there is any missing, delete the target student’s period superimposed features and period independent features in the current period.

[0049] To solve the above technical problems, the third aspect of the present invention provides an electronic device, comprising: at least one processor; and a memory, wherein the memory stores instructions, and when the instructions are executed by the at least one processor, the at least one processor executes the integrated learning performance prediction method as described in any of the above embodiments.

[0050] To solve the above technical problems, the fourth aspect of the present invention provides a machine-readable storage medium storing executable instructions, which, when executed, enable the machine to execute the integrated learning performance prediction method as described in any of the above embodiments.

[0051] Different from the prior art, the beneficial effects of this application are:

[0052] This application uses the XGBoost model and one-hot encoding to perform binary processing on the students' learning behavior data, and then inputs it into the soft voting classifier. The soft voting classifier fuses the XGBoost model, the logistic regression (LR) model, and the random forest (RF) model. The fusion prediction result is obtained by weighted averaging the prediction data of the three models, benefiting from the powerful boost of XGBoost, the simplicity of LR, and the integration strength of RF, thereby solving the problem of poor prediction performance caused by the single lowest-level model of the traditional prediction method, and effectively improving the prediction performance and accuracy;

[0053] This application obtains the iterative information of learning behavior data based on the XGBoost model, and obtains a binary feature vector after one-hot encoding the iterative information. Performance prediction and model training are performed based on the binary feature vector, which effectively improves the expressiveness and segmentation accuracy of the features, improves the training effect of the soft voting classifier, and improves the model prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flowchart of an implementation method of the integrated learning performance prediction method of the present application;

[0055] Figure 2 yes Figure 1 A schematic flow chart of an implementation method corresponding to S100;

[0056] Figure 3 It is a flowchart of an implementation method of the XGBoost model training method of the present application;

[0057] Figure 4 It is a schematic diagram of an implementation method of one-hot encoding of the present application;

[0058] Figure 5 It is a flowchart of an implementation method of a training method of a soft voting classifier of the present application;

[0059] Figure 6 It is the prediction effect experimental data diagram of this application;

[0060] Figure 7 This is a structural diagram of an implementation method of an integrated learning performance prediction device of the present application;

[0061] Figure 8 It is a schematic structural diagram of an embodiment of an electronic device of the present application. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described 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, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0063] At present, there have been many studies related to machine learning performance prediction, but they mainly use a single model for prediction, or use stacking structure models for prediction. The former single model has poor prediction performance and certain limitations when facing the complex and changeable behavioral data of students in the online and offline hybrid teaching mode. Although the latter uses multiple models, the output of the previous model is directly used as the input of the next model, and the lowest-level meta-learner still cannot effectively guarantee the processing performance of complex data.

[0064] In order to solve the above problems, the applicant has developed an integrated learning performance prediction method, which uses a multi-layer model to predict students' grades based on their online and offline learning behavior data within a certain period. Through new model fusion and training methods, the prediction performance can be significantly improved. It can effectively predict grades based on students' behavior data within a shorter period, which helps teachers to understand students' learning situation more promptly and take targeted measures, and can also improve students' self-consciousness in learning to a certain extent.

[0065] For details, please refer to Figure 1 , Figure 1 It is a flow chart of an implementation method of the integrated learning performance prediction method of the present application.

[0066] like Figure 1 As shown, the method includes:

[0067] S100. Obtain learning behavior data of target students.

[0068] Among them, the learning behavior data may include online learning behavior data and offline learning behavior data within a certain period.

[0069] The certain period can be a study week, a month, a quarter, or any other preset period, which can achieve the effect of the present embodiment. In order to ensure that sufficient data can be provided and improve the timeliness of the performance prediction, the certain period is preferably a study week.

[0070] Online learning behavior data is used to record students' online learning situation. For example, online learning behavior data may include: number of completed tasks, course video progress, video viewing time, number of chapter learning times, chapter test scores, course points, etc. Correspondingly, offline learning behavior data is used to record students' offline learning situation. For example, offline learning behavior data may include: number of discussions participated in, discussion scores, homework scores, attendance rates, student interaction scores, group task scores, number of votes, number of electors, and in-class test scores.

[0071] Specifically, the learning behavior data may be shown in the following Table 1:

[0072] Table 1

[0073]

[0074] In one embodiment, the learning behavior data of all students can be defined as: ,in Are a student m In the n The set of learning behavior characteristics of a cycle can be defined as: ,in Are a student m In the n The first cycle t learning behavior features, each learning behavior feature is a piece of behavior data obtained.

[0075] It can be understood that in this embodiment, the learning behavior data may include both online learning behavior data and offline learning behavior data. In other embodiments, the learning behavior data may also include only online or offline learning behavior data.

[0076] In one embodiment, learning behavior data can be divided into cycle-superimposed features and cycle-independent features, wherein cycle-superimposed features are features that gradually increase with the increase of student learning cycles, such as the number of completed courses, course video progress, etc., while cycle-independent features are features completed by students in the current cycle, such as homework scores and in-class test scores.

[0077] Therefore, in order to obtain the period superposition characteristics of students, it is necessary to subtract the student's data in the current week from the data in the previous week. The following describes in detail how to obtain the student's learning behavior data. Figure 2 , Figure 2 yes Figure 1 A flow chart of an implementation method corresponding to S100 in FIG.

[0078] like Figure 2 As shown, the method for obtaining learning behavior data includes:

[0079] S101. Obtain the cycle-independent features of the target student in the current cycle, and the superposition value of the cycle-superposition features in the current cycle and the previous cycle.

[0080] S102. Obtain the period superposition feature of the target student in the current period based on the superposition value of the period superposition feature of the target student in the current period and the previous period.

[0081] Among them, the superposition value of the period superposition feature is the cumulative data of the period superposition feature. For example, for the number of course completions, the total number of course completions a of the target student in the current cycle and the total number of course completions b of the previous cycle can be obtained, thereby obtaining the number of course completions ab of the target student in the current cycle.

[0082] S103, determining whether the target student's period superposition characteristics and period independent characteristics in the current period have negative values ​​or are missing.

[0083] S104. If there is a negative value, re-extract the corresponding student data.

[0084] S105. If there is any missing, delete the period superposition features and period independent features of the target student in the current period.

[0085] In order to obtain periodic superposition features, some features may have negative values ​​when the data of adjacent periods are subtracted. At the same time, there may be missing data when the data is acquired, so data preprocessing is required.

[0086] When there are negative values, the corresponding student data can be extracted again; when the data is missing, the student's data can be deleted.

[0087] S200, using a pre-trained XGBoost model to predict iteration information of learning behavior data during the XGBoost model training process.

[0088] The iteration information includes the leaf nodes that the learning behavior data may fall into in all decision trees iteratively generated during the XGBoost model training process.

[0089] The XGBoost model (eXtreme Gradient Boosting) is an efficient gradient boosting decision tree algorithm. It optimizes the traditional GBDT (Gradient Boosting Decision Tree) to improve the prediction accuracy and execution efficiency of the model.

[0090] This application uses a pre-trained XGBoost model as the first-layer model of the prediction method. During the training process, the XGBoost model continuously iterates to generate decision trees to fit the residual between the prediction result of the previous tree and the true value until the preset number of iterations is reached or the loss is no longer significantly reduced. t The formula for fitting the residual of the previous tree is:

[0091] .

[0092] Specifically, during the training process of the XGBoost model, the decision tree generated in each iteration can be recorded. Since the decision tree is known, the trained XGBoost model can be used to predict the leaf nodes that the learning behavior data may fall into in the decision tree during its iteration, that is, the iteration information.

[0093] The following is a detailed description of the training method of the XGBoost model in this application. Figure 3 , Figure 3 It is a flowchart of an implementation method of the XGBoost model training method of the present application.

[0094] like Figure 3 As shown, the training method of the XGBoost model includes:

[0095] S201, obtaining a sample target set and dividing it into a sample training set and a sample test set.

[0096] Among them, the sample target set includes learning behavior data with labeled sample labels.

[0097] In one embodiment, the sample label may include 0 and 1, 0 represents an unsatisfactory final exam score, and 1 represents a qualified final exam score; in other embodiments, the sample labels may also be designed based on actual needs. For example, the sample label may include three, representing unsatisfactory, qualified, excellent, etc., which can achieve the effect of this embodiment.

[0098] S202. Based on the grid search method, the XGBoost model is cross-validated with the sample training set to evaluate the performance of each set of parameters, and then the optimal model parameters are obtained based on the sample test set.

[0099] The grid search method exhaustively enumerates all possible parameter combinations in a preset parameter grid, then uses a sample training set to cross-validate the parameter combinations, conducts performance evaluation, and finally obtains the optimal model parameters through a sample test set.

[0100] In one embodiment, the cross validation may be specifically a five-fold cross validation, that is, the data is divided into five parts, and training and validation are performed separately. In other embodiments, other types of cross validation may also be used, and the effect of this embodiment can be achieved.

[0101] S300, performing one-hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data.

[0102] After S200 obtains the iterative information of the target student's learning behavior data, the iterative information can be further one-hot encoded. One-hot encoding can improve the feature's expressiveness and segmentation accuracy, and convert the original feature data into a form that is easier for the prediction model to understand and process. Therefore, the feature matrix processed by feature encoding is often better interpretable than the original feature matrix during the prediction model operation process.

[0103] In addition, the one-hot encoding in this embodiment is not applied to the centralized preprocessing process of the data. Therefore, the one-hot encoding process will not affect the influence of the original feature data on the model, and the extracted data itself can be retained and its role can be maximized.

[0104] The following is a detailed description of the one-hot encoding method. Figure 4 , Figure 4 Schematic diagram of one implementation method of the unique hot encoding of the present application. Figure 4 As shown, the one-hot encoding method can be to arrange all decision trees iteratively generated during the XGBoost model training process in parallel in the order of generation, and then set the leaf nodes where the learning behavior data falls in the decision tree to one, and set the other leaf nodes to zero, to obtain a binary feature vector of the learning behavior data.

[0105] S400, inputting the binary feature vector into a pre-trained soft voting classifier to obtain a predicted result of the target student's score.

[0106] The soft voting classifier includes a parallel XGBoost model, an LR model, and a RF model to perform weighted averaging on the prediction data of the XGBoost model, the LR model, and the RF model.

[0107] This application integrates the XGBoost model, logistic regression (LR) model and random forest (RF) model in the second-level prediction, and obtains the integrated prediction result by inputting the prediction data of the three models into the weight model at the same time, benefiting from the powerful boost of XGBoost, the simplicity of LR and the integration strength of RF, thereby solving the problem of poor prediction performance caused by the single bottom-level model of traditional prediction methods.

[0108] In one embodiment, the weighted average formula may be as follows:

[0109] ;

[0110] In the formula, is the weight of each algorithm’s prediction data, is the predicted probability of each algorithm predicting the data.

[0111] The following is a detailed description of the training method of the soft voting classifier in this application. Figure 5 , Figure 5 It is a flowchart of an implementation method of a training method of a soft voting classifier of the present application.

[0112] like Figure 5 As shown, the training method includes:

[0113] S401, recording the first training iteration information of the sample training set during the XGBoost model training process.

[0114] The first training iteration information includes the leaf nodes into which each learning behavior data in the sample training set falls in all decision trees iteratively generated during the XGBoost model training process.

[0115] S402: Use the trained XGBoost model to predict the second training iteration information of the sample test set during the XGBoost model training process.

[0116] The second training iteration information includes the leaf nodes that each learning behavior data in the sample test set may fall into in all decision trees iteratively generated during the XGBoost model training process.

[0117] Similar to S200, since the XGBoost model continuously iterates to generate a decision tree during the training process, each input learning behavior data will fall into a leaf node in the decision tree of each iteration. By recording this information, the first training iteration information of the sample training set can be obtained.

[0118] Since the iteratively generated decision tree has been recorded, the trained XGBoost model can be used to predict the leaf nodes that the learning behavior data in the sample test set falls into in the decision tree during its iteration, that is, the second training iteration information.

[0119] S403 , performing one-hot encoding on the first training iteration information and the second training iteration information to obtain a binary feature vector of each learning behavior data.

[0120] In this implementation, the first training iteration information of the sample training set and the second training iteration information of the sample test set are one-hot encoded, thereby improving the training effect of the soft voting classifier and improving the model prediction accuracy.

[0121] Similar to S300, the one-hot encoding method can be to arrange all decision trees iteratively generated during the XGBoost model training process in parallel in the order of generation, and then set the leaf nodes where the target learning behavior data in the decision tree falls to one, and set the other leaf nodes to zero, to obtain a binary feature vector of the target learning behavior data.

[0122] S404: Based on the binary feature vectors of the learning behavior data and the labeled sample labels thereof, a binary training set and a binary test set are obtained.

[0123] Based on one-hot encoding, a binary feature vector corresponding to each learning behavior data can be obtained. Furthermore, based on the sample labels annotated with the learning behavior data, the sample labels can be annotated on the binary feature vector to obtain a binary training set and a binary test set.

[0124] Among them, the binary training set corresponds to the sample training set in the XGBoost training process, and the binary test set corresponds to the sample test set in the XGBoost training process.

[0125] S405: Use the binary training set and the binary test set to train and test the soft voting classifier.

[0126] Specifically, in one embodiment, similar to the training method of the XGBoost model, the soft voting classifier can also be trained based on the grid search method, and the soft voting classifier is cross-validated with a binary training set to evaluate the performance of each set of parameters, and then the optimal model parameters of the three models under the soft voting classifier are obtained based on the binary test set.

[0127] The cross-validation may be a five-fold cross-validation or other types of cross-validation, both of which can achieve the effect of the present embodiment.

[0128] In other implementations, the soft voting classifier may also be trained based on other commonly used training methods, such as training based on a loss function until the loss function converges, which can achieve the effect of this implementation.

[0129] Based on the above training process, by using the above binary training set and binary test set to train the soft voting classifier, the soft voting classifier's understanding of the data can be significantly improved, thereby improving the training effect and the prediction effect.

[0130] In order to test the effect of the XGBoost+LR+RF score prediction model of the present application, the applicant used the same sample test set to test the XGBoost+LR+RF score prediction model of the present application and the existing 8 machine learning algorithms. The optimal parameters of each model were found by the grid search method. The prediction results of each model under the optimal parameters are shown in Table 2 below. At the same time, the ROC curve and the area under the curve AOC were calculated to obtain Figure 6 , Figure 6 It is a data diagram of the prediction effect experiment of this application.

[0131] Table 2

[0132]

[0133] See Table 2 and Figure 6 data, Figure 6 The closer the ROC curve is to the upper left corner, that is, the larger the lower right area is, the better the model's ability to distinguish between positive and negative samples is. Therefore, it can be seen that the XGBoost+LR+RF score prediction model of this application performs well in both accuracy and AUC value, with an accuracy of 96.18% and an AUC value of 98.7%, which is of great significance to the realization of the score warning system under the hybrid teaching mode.

[0134] This application also provides an integrated learning performance prediction device, see Figure 7 , Figure 7 It is a structural diagram of an implementation method of an integrated learning performance prediction device of the present application.

[0135] like Figure 7 As shown, the integrated learning performance prediction device includes a data acquisition module 21, an iteration information acquisition module 22, an encoding module 23 and a fusion prediction module 24.

[0136] The data acquisition module 21 is used to acquire the learning behavior data of the target students, and the learning behavior data includes online learning behavior characteristics and / or offline learning behavior characteristics within a certain period;

[0137] The iteration information acquisition module 22 is used to use the pre-trained XGBoost model to predict the iteration information of the learning behavior data during the XGBoost model training process, and the iteration information includes the leaf nodes that the learning behavior data may fall into in all decision trees iteratively generated during the XGBoost model training process;

[0138] The encoding module 23 is used to perform one-hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data;

[0139] The fusion prediction module 24 is used to input the binary feature vector into a pre-trained soft voting classifier to obtain the target student's score prediction result. The soft voting classifier includes a parallel XGBoost model, an LR model and an RF model to perform weighted averaging on the prediction data of the XGBoost model, the LR model and the RF model.

[0140] In one embodiment, the integrated learning performance prediction device also includes a first model training module 25, which is used to obtain a sample target set and divide it into a sample training set and a sample test set, wherein the sample target set includes learning behavior data with labeled sample labels; based on the grid search method, the XGBoost model is cross-validated with the sample training set to evaluate the performance of each set of parameters, and then the optimal model parameters are obtained based on the sample test set.

[0141] In one embodiment, the integrated learning performance prediction device also includes a second model training module 26.

[0142] The second model training module 26 is used to record the first training iteration information of the sample training set during the XGBoost model training process, the first training iteration information includes the leaf nodes that each learning behavior data in the sample training set falls into in all decision trees iteratively generated during the XGBoost model training process; use the trained XGBoost model to predict the second training iteration information of the sample test set during the XGBoost model training process, the second training iteration information includes the leaf nodes that each learning behavior data in the sample test set may fall into in all decision trees iteratively generated during the XGBoost model training process; perform one-hot encoding on the first training iteration information and the second training iteration information to obtain a binary feature vector of each learning behavior data; obtain a binary training set and a binary test set based on the binary feature vector of the learning behavior data and its annotated sample label; and use the binary training set and the binary test set to train the soft voting classifier.

[0143] As above Figures 1 to 6 , the integrated learning performance prediction method according to the embodiment of this specification is described. The details mentioned in the above description of the method embodiment are also applicable to the integrated learning performance prediction device of the embodiment of this specification. The above integrated learning performance prediction device can be implemented by hardware, or by software, or a combination of hardware and software.

[0144] The present application also provides an electronic device, see Figure 8 , Figure 8 Schematic diagram of the structure of an electronic device of the present application. Figure 8As shown, the electronic device 30 may include at least one processor 31, a memory 32 (e.g., a non-volatile memory), a memory 33, and a communication interface 34, and the at least one processor 31, the memory 32, the memory 33, and the communication interface 34 are connected together via a bus 35. At least one processor 31 executes at least one computer-readable instruction stored or encoded in the memory 32.

[0145] It should be understood that the computer executable instructions stored in the memory 32, when executed, cause at least one processor 31 to perform the above combined operations in various embodiments of the present specification. Figure 1-Figure 5 Describes the various operations and functions.

[0146] In the embodiments of the present specification, the electronic device 30 may include, but is not limited to, personal computers, server computers, workstations, desktop computers, laptop computers, notebook computers, mobile electronic devices, smart phones, tablet computers, cellular phones, personal digital assistants (PDAs), handheld devices, messaging devices, wearable electronic devices, consumer electronic devices, and the like.

[0147] According to one embodiment, a program product such as a machine-readable medium is provided. The machine-readable medium may have instructions (i.e., the above-mentioned elements implemented in the form of software), which, when executed by a machine, causes the machine to perform the above-mentioned combination of various embodiments of this specification. Figure 1-Figure 7 Specifically, a system or device equipped with a readable storage medium may be provided, on which a software program code implementing the functions of any of the above-mentioned embodiments is stored, and a computer or processor of the system or device reads and executes the instructions stored in the readable storage medium.

[0148] In this case, the program code itself read from the machine-readable medium can realize the function of any one of the above embodiments, and thus the machine-readable code and the machine-readable storage medium storing the machine-readable code constitute part of this specification.

[0149] Examples of readable storage media include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code may be downloaded from a server computer or a cloud via a communication network.

[0150] Those skilled in the art should understand that the various embodiments disclosed above can be modified and altered in various ways without departing from the essence of the invention. Therefore, the protection scope of this specification should be defined by the appended claims.

[0151] It should be noted that not all steps and units in the above-mentioned processes and system structure diagrams are necessary, and some steps or units can be ignored according to actual needs. The execution order of each step is not fixed and can be determined as needed. The device structure described in the above-mentioned embodiments can be a physical structure or a logical structure, that is, some units may be implemented by the same physical client, or some units may be implemented by multiple physical clients, or some components in multiple independent devices may be implemented together.

[0152] In the above embodiments, the hardware unit or module can be realized by mechanical or electrical means. For example, a hardware unit, module or processor can include permanent dedicated circuit or logic (such as special processor, FPGA or ASIC) to complete the corresponding operation. The hardware unit or processor can also include programmable logic or circuit (such as general-purpose processor or other programmable processor), which can be temporarily set by software to complete the corresponding operation. Specific implementation (mechanical method or dedicated permanent circuit or temporary circuit) can be determined based on cost and time consideration.

[0153] The specific embodiments described above in conjunction with the accompanying drawings describe exemplary embodiments, but do not represent all embodiments that can be implemented or fall within the scope of protection of the claims. The term "exemplary" used throughout this specification means "used as an example, instance or illustration" and does not mean "preferred" or "having advantages" over other embodiments. For the purpose of providing an understanding of the described technology, the specific embodiments include specific details. However, these technologies can be implemented without these specific details. In some instances, in order to avoid making the concepts of the described embodiments difficult to understand, well-known structures and devices are shown in block diagram form.

[0154] The above description of the present disclosure is provided to enable any person of ordinary skill in the art to implement or use the present disclosure. Various modifications to the present disclosure will be apparent to those of ordinary skill in the art, and the general principles corresponding to the present disclosure may be applied to other variations without departing from the scope of protection of the present disclosure. Therefore, the present disclosure is not limited to the examples and designs described herein, but is consistent with the widest range of principles and novel features disclosed herein.

Claims

1. An integrated learning performance prediction method, characterized in that: include: Acquire learning behavior data of target students, wherein the learning behavior data includes online learning behavior characteristics and / or offline learning behavior characteristics within a certain period; Using a pre-trained XGBoost model to predict iteration information of the learning behavior data during the XGBoost model training process, the iteration information includes leaf nodes that the learning behavior data may fall into in all decision trees iteratively generated during the XGBoost model training process; One-hot encoding the iteration information to obtain a binary feature vector of the learning behavior data; Inputting the binary feature vector into a pre-trained soft voting classifier to obtain a predicted result of the target student's score, wherein the soft voting classifier includes the XGBoost model, the LR model, and the RF model in parallel to perform weighted averaging on the predicted data of the XGBoost model, the LR model, and the RF model; Wherein, the learning behavior characteristics include period superposition characteristics and period independence characteristics; The step of obtaining the learning behavior data of the target student includes: Obtaining the period-independent features of the target student in the current period, and the superposition value of the period-superposition features in the current period and the previous period, and obtaining the period-superposition features of the target student in the current period based on the superposition value of the period-superposition features of the target student in the current period and the previous period; Determine whether the target student's period superposition characteristics and period independent characteristics in the current period have negative values ​​or are missing; If there are negative values, re-extract the corresponding student data; If there is a missing, delete the target student's period superposition characteristics and period independent characteristics in the current period; The step of performing one-hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data comprises: Arrange all decision trees iteratively generated during the XGBoost model training process in parallel according to the generation order; The leaf node where the learning behavior data falls in the decision tree is set to 1, and the other leaf nodes are set to 0, to obtain a binary feature vector of the learning behavior data.

2. The ensemble learning performance prediction method according to claim 1, characterized in that: The training method of the XGBoost model includes: Acquire a sample target set and divide it into a sample training set and a sample test set, wherein the sample target set includes the learning behavior data with labeled sample labels; Based on the grid search method, the XGBoost model is cross-validated with the sample training set to evaluate the performance of each set of parameters, and then the optimal model parameters are obtained based on the sample test set.

3. The ensemble learning performance prediction method according to claim 2, characterized in that: The training method of the soft voting classifier includes: Recording the first training iteration information of the sample training set during the XGBoost model training process, wherein the first training iteration information includes the leaf nodes into which each learning behavior data in the sample training set falls in all decision trees iteratively generated during the XGBoost model training process; Use the trained XGBoost model to predict the second training iteration information of the sample test set in the XGBoost model training process, wherein the second training iteration information includes the leaf nodes that each learning behavior data in the sample test set may fall into in all decision trees iteratively generated during the XGBoost model training process; Performing one-hot encoding on the first training iteration information and the second training iteration information to obtain a binary feature vector of each of the learning behavior data; Based on the binary feature vectors of the learning behavior data and the labeled sample labels thereof, a binary training set and a binary test set are obtained; The soft voting classifier is trained and tested using the binary training set and the binary test set.

4. The ensemble learning performance prediction method according to claim 3, characterized in that: The step of training and testing the soft voting classifier using the binary training set and the binary test set comprises: Based on the grid search method, the soft voting classifier is cross-validated with the binary training set to evaluate the performance of each set of parameters, and then the optimal parameters of the three models under the soft voting classifier are obtained based on the binary test set.

5. The ensemble learning performance prediction method according to claim 2, characterized in that: The sample labels include 0 and 1, 0 represents a failed final exam result, and 1 represents a passed final exam result; and / or, The learning behavior data include online learning behavior characteristics and offline learning behavior characteristics within a certain period. The online learning behavior data include the number of tasks completed, course video progress, video viewing time, number of chapter learning times, chapter test scores, and course points. The offline learning behavior characteristics include the number of discussions participated in, discussion scores, homework scores, sign-in rates, student interaction scores, group task scores, number of votes, number of selected people, and in-class test scores.

6. An integrated learning performance prediction device, used to implement the integrated learning performance prediction method according to any one of claims 1 to 5, characterized in that: include: A data acquisition module, used to acquire the learning behavior data of the target students, wherein the learning behavior data includes online learning behavior characteristics and / or offline learning behavior characteristics within a certain period; An iteration information acquisition module, used to use a pre-trained XGBoost model to predict the iteration information of the learning behavior data during the training process of the XGBoost model, wherein the iteration information includes the leaf nodes that the learning behavior data may fall into in all decision trees iteratively generated during the training process of the XGBoost model; An encoding module, used for performing one-hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data; A fusion prediction module, used for inputting the binary feature vector into a pre-trained soft voting classifier to obtain the score prediction result of the target student, wherein the soft voting classifier includes the XGBoost model, the LR model and the RF model in parallel to perform weighted averaging on the prediction data of the XGBoost model, the LR model and the RF model; Wherein, the learning behavior characteristics include period superposition characteristics and period independence characteristics; The step of obtaining the learning behavior data of the target student includes: Obtaining the cycle-independent features of the target student in the current cycle and the superposition value of the cycle-superposition features in the current cycle and the previous cycle, and obtaining the cycle-superposition features of the target student in the current cycle based on the superposition value of the cycle-superposition features of the target student in the current cycle and the previous cycle; Determine whether the target student's period superposition characteristics and period independent characteristics in the current period have negative values ​​or are missing; If there are negative values, re-extract the corresponding student data; If there is a missing, delete the target student's period superposition characteristics and period independent characteristics in the current period; The step of performing one-hot encoding on the iteration information to obtain a binary feature vector of the learning behavior data comprises: Arrange all decision trees iteratively generated during the XGBoost model training process in parallel according to the generation order; The leaf node where the learning behavior data falls in the decision tree is set to 1, and the other leaf nodes are set to 0, to obtain a binary feature vector of the learning behavior data.

7. An electronic device comprising: at least one processor; and a memory storing instructions, which, when executed by the at least one processor, causes the at least one processor to execute the integrated learning performance prediction method as described in any one of claims 1 to 5.

8. A machine-readable storage medium storing executable instructions, which, when executed, enable the machine to perform the integrated learning performance prediction method according to any one of claims 1 to 5.

Images