Hierarchical structure score prediction method and device, equipment and medium

Abstract

The invention discloses a score prediction method and device of a hierarchical structure, equipment and a medium, and relates to the technical field of big data. The score prediction method comprises steps S1 to S4. The method comprises the steps of S1, obtaining training data, and performing cleaning and preprocessing; wherein the training data comprises a plurality of feature data of a plurality of students. Student scores are divided into a plurality of score categories. And S2, according to the training data, constructing a student score prediction model comprising a multi-layer binary classifier. Wherein the multiple layers of binary classifiers which are connected in sequence are respectively used for judging whether the student scores belong to score categories of which the student distribution is ranked from more to less. And S3, respectively training each layer of binary classifier of the student score prediction model by adopting cross validation and grid search according to the training data to obtain a trained student score prediction model. And S4, obtaining student data to be predicted, and inputting the student data to the trained student score prediction model to obtain the predicted score of the student.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a hierarchical performance prediction method, device, equipment and medium. Background Art

[0002] Mining hidden patterns in education data to predict student performance can provide objective reference for education policy, promote systematic improvement of the education environment, and enable parents to better support their children's education.

[0003] At present, most student performance prediction models use a method of directly performing multi-classification prediction on performance. Specifically, the performance-related data is first collected, and then divided into training and test sets after cleaning and preprocessing. The model is trained with the training set and the optimal parameters are obtained through grid search. Finally, the prediction performance of the algorithm is verified with the test set. However, the category distribution of actual student performance data sets is often unbalanced. For example, the proportion of excellent, good, medium, passing, and failing students varies greatly. This causes the prediction results of existing models to be biased towards the majority class and perform poorly when detecting the minority class. Common sampling techniques for dealing with imbalanced problems, such as oversampling, which may introduce bad samples, and undersampling, which may remove important samples, all have certain defects. Summary of the invention

[0004] The present invention provides a hierarchical performance prediction method, device, equipment and medium to improve at least one of the above technical problems.

[0005] In a first aspect, the present invention provides a hierarchical performance prediction method, which comprises steps S1 to S4.

[0006] S1. Obtain training data, and perform cleaning and preprocessing. The training data includes multiple feature data of multiple students. The student scores are divided into multiple score categories.

[0007] S2. Based on the training data, a student score prediction model including a multi-layer CatBoost binary classifier is constructed, wherein the multi-layer CatBoost binary classifiers connected in sequence are respectively used to determine whether the student score belongs to each score category arranged from most to least.

[0008] S3. Using cross-validation and grid search according to the training data, train each layer of CatBoost binary classifiers of the student performance prediction model respectively to obtain a trained student performance prediction model.

[0009] S4. Obtain the student data to be predicted, and input the trained student performance prediction model to obtain the student's predicted performance.

[0010] In an optional implementation, step S1 specifically includes steps S11 to S14.

[0011] S11, obtaining training data. The training data includes multiple feature data of multiple students. The features of the feature data include academic performance, demographic information, social background data and school-specific features. The academic performance includes the school where the student is located, the weekly study time, the number of past course failures, the first semester grades, the second semester grades, and the final grades. The demographic information includes the student's gender, the student's age, the student's nationality, and whether the student is an international student. The social background data includes free time after school, extracurricular activities, the type of student's home address, the father's education level, the mother's education level, the father's occupation, the mother's occupation, and the time from home to school. The school-specific features include the reason for choosing this school, the number of school absences, whether the student has any special educational needs, the number of course units approved by the student in the first semester, the number of course units approved by the student in the second semester, the average grade in the first semester, and the average grade in the second semester.

[0012] S12. Clean and preprocess the training data, and obtain the feature importance of each feature to the learning performance through a random forest model with default hyperparameters.

[0013] S13. Filter out useful features according to the feature importance and obtain useful training data.

[0014] S14. According to the useful training data, the student scores are divided into a plurality of score categories to obtain a final training data set.

[0015] In an optional implementation, when the grade value of the learning performance is between 0 and 100, the performance category is determined by the following thresholds: If the value is less than 60, it is classified as “failure”; If the value is between 60 and 69, inclusive, it is classified as “qualified”; If the value is between 70 and 79, inclusive, it is classified as “satisfactory”; If the value is between 80 and 89, inclusive, it is classified as “good”; If the value is 90 or above, it is classified as "excellent".

[0016] In an optional implementation, step S2 specifically includes steps S21 to S23.

[0017] S21. Obtain the number n of grades of the student's grades.

[0018] S22. According to the number of score categories, a student score prediction model is constructed by sequentially connecting n-1 CatBoost binary classifiers. The sequentially connected n-1 CatBoost binary classifiers are used to predict whether the student score belongs to the score category with the most students, whether it belongs to the score category with the second most students, whether it belongs to the score category with the third most students, and so on. The last CatBoost binary classifier is used to predict whether the student score belongs to the score category with the second least students.

[0019] S23. Among the n-1 CatBoost binary classifiers connected in sequence in the student score prediction model, when one CatBoost binary classifier predicts that the student score belongs to a certain score category, the prediction result is output; otherwise, the next layer of CatBoost binary classifiers continues to predict whether the student score belongs to the score category of the next student distribution, until the last layer of CatBoost binary classifiers completes the prediction and outputs the result of the last layer.

[0020] In an optional implementation, step S3 specifically includes step S31 to step S32.

[0021] S31. Based on the training data, 10-fold cross validation is performed: first, the data set is divided into 10 parts. In each iteration, one of the parts is used as a test set to verify the model performance, and the remaining 9 parts are used as training sets to train the model. This process is repeated 10 times, and the training set and test set are replaced each time. Finally, the average of the 10 results is used as the final performance evaluation of the model.

[0022] S32. Each time cross-validation is performed, the grade category with the most students in the training set is first divided into one category, and the remaining grade categories are divided into one category, and the first layer of CatBoost binary classifiers are trained. After completing the training of a CatBoost binary classifier, the training data of the grade category with the most students is deleted, and the grade category with the most students in the training data after the deletion is divided into one category, and the other grade categories are divided into one category, and the CatBoost binary classifier of the next layer is continued to be trained until the training of the last layer of CatBoost binary classifier is completed. Wherein, during training, the optimal binary classifier of the current layer is obtained based on grid search.

[0023] In an optional implementation, the optimal binary classifier of the current layer is obtained based on a grid search, specifically including: CatBoost binary classifier needs to adjust two hyperparameters, the number of base classifiers and the depth of each tree. Each hyperparameter has three candidate values, the candidate values ​​of the number of base classifiers are 100, 300 and 500, and the candidate values ​​of the depth of each tree are 3, 4 and 5. Among them, the candidate values ​​of the number of base classifiers and the candidate values ​​of the depth of each tree are randomly combined to obtain 9 pairs of hyperparameters [100,3], [100,4], [100,5], [300,3], [300,4], [300,5], [500,3], [500,4] and [500,5]. All these 9 pairs of hyperparameters are used for model training, and the best hyperparameter set is selected according to its performance.

[0024] In an optional implementation, step S4 specifically includes step S41 to step S42.

[0025] S41. Obtain the student data to be predicted.

[0026] S42. Input the student data to be predicted into the trained student score prediction model, and use the multi-layer CatBoost binary classifier of the student score prediction model to predict layer by layer whether it belongs to the score category corresponding to the current layer. When it is predicted to belong to the score category corresponding to the current layer, the prediction result is output; otherwise, the data is input into the next layer for further judgment until the score category corresponding to the student data to be predicted is determined.

[0027] In a second aspect, the present invention provides a hierarchical performance prediction device, which includes a training data acquisition module, a model construction module, a model training module and an inference module.

[0028] The training data acquisition module is used to acquire, clean and pre-process the training data. The training data includes multiple feature data of multiple students. The student scores are divided into multiple score categories. The model building module is used to build a student score prediction model including multiple layers of CatBoost binary classifiers according to the training data. The multiple layers of CatBoost binary classifiers connected in sequence are used to determine whether the student score belongs to each score category arranged from most to least students. The model training module is used to train each layer of CatBoost binary classifiers of the student score prediction model by cross-validation and grid search according to the training data to obtain a trained student score prediction model.

[0029] The inference module is used to obtain the student data to be predicted and input the trained student performance prediction model to obtain the student's predicted performance.

[0030] In a third aspect, the present invention provides a hierarchical performance prediction device, comprising a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a hierarchical performance prediction method as described in any paragraph of the first aspect.

[0031] In a fourth aspect, the present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute a hierarchical performance prediction method as described in any paragraph of the first aspect.

[0032] By adopting the above technical solution, the present invention can achieve the following technical effects: A hierarchical performance prediction method of the present invention effectively solves the problem of class imbalance in educational data mining and significantly improves the accuracy of the prediction model. The method constructs a multi-layer CatBoost model and processes data layer by layer in a refined manner, so that the model can deeply capture the complex patterns and nonlinear relationships behind students' performance. In addition, the technical solution optimizes large-scale data processing capabilities, and sample optimization at each layer reduces the computational burden while maintaining an efficient computing speed. More importantly, the generalization ability of the method enables it to adapt to different student groups and educational environments, providing accurate data support for teachers and school administrators, helping them identify students who need additional help, and realizing the rational allocation of educational resources and the formulation of personalized teaching strategies. Through early intervention and targeted teaching support, this patented technical solution is expected to significantly improve educational effectiveness and promote students' academic success. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the specific implementation methods of the present invention. It should be understood that the following drawings only show certain specific implementation methods of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 It is a flowchart of a hierarchical performance prediction method. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] Example 1, please refer to Figure 1 The first embodiment of the present invention provides a hierarchical performance prediction method, which can be performed by a hierarchical performance prediction device (hereinafter referred to as: performance prediction device). In particular, it is performed by one or more processors in the performance prediction device to implement steps S1 to S4.

[0037] S1. Obtain training data, and perform cleaning and preprocessing. The training data includes multiple feature data of multiple students. The student scores are divided into multiple score categories. Preferably, step S1 specifically includes steps S11 to S14.

[0038] S11, obtaining training data, wherein the training data includes multiple characteristic data of multiple students, and the characteristics of the characteristic data include academic performance, demographic information, social background data and school-specific characteristics.

[0039] The academic performance includes the school the student is attending, the hours studied per week, the number of failed courses in the past, the first semester grades, the second semester grades, and the final grades.

[0040] The demographic information includes student gender, student age, student nationality, and whether the student is an international student.

[0041] The social background data included free time after school, extracurricular activities, type of student home address, father's education level, mother's education level, father's occupation, mother's occupation, and time from home to school.

[0042] The school-specific characteristics described included the reason for choosing this school, the number of absences from school, whether the student had any special educational needs, the number of course units approved by the student in the first semester, the number of course units approved by the student in the second semester, the average grade in the first semester, and the average grade in the second semester.

[0043] S12. Clean and preprocess the training data, and obtain the feature importance of each feature to the learning performance through a random forest model with default hyperparameters.

[0044] S13. Filter out useful features according to the feature importance and obtain useful training data.

[0045] Among many features, some features are more relevant to the category to be predicted, while others are completely irrelevant. Screening out the features with higher relevance in advance and removing the features with lower relevance can effectively improve the accuracy of prediction. In this embodiment, the RF (random forest) model with default hyperparameters is used to calculate each feature, thereby obtaining the relevance of each feature (i.e., feature importance).

[0046] For example, the four most important features for predicting student dropout and academic success are the number of units approved by the student in the second semester, the number of units approved by the student in the first semester, the second semester average grade, and the first semester average grade. This is because completing a course is critical to academic progress and success. The higher the numbers for these features, the more likely the student is to succeed in the course, meet the requirements, and get closer to graduation.

[0047] Additionally, the average of grades for Semester 1 and Semester 2 can provide an indication of a student's academic performance over time. Consistently poor performance (low grades) may indicate underlying issues such as learning difficulties, lack of engagement or personal challenges that may lead to dropout.

[0048] Nationality (the student’s nationality), internationality (whether the student is an international student), and educational needs (whether the student has any special educational needs) are the least important characteristics for the task of academic progress and success. This may be because the main focus of predicting dropout and academic success is on academic factors, such as grades, course completion rates, and overall academic performance. These indicators directly reflect the student’s level of engagement with the coursework, understanding of the material, and ability to make academic progress. In contrast, factors such as nationality, international status, and educational needs are generally not directly related to these academic factors.

[0049] S14. According to the useful training data, the student scores are divided into a plurality of score categories to obtain a final training data set.

[0050] When the grade value of the academic performance is between 0 and 150, the performance category is determined by the following thresholds: If the value is less than 69, it is classified as "unsatisfactory". If the value is between 70 and 89, inclusive, it is classified as "pass". If the value is between 90 and 109, inclusive, it is classified as "satisfactory". If the value is between 110 and 129, inclusive, it is classified as "good". If the value is 130 or above, it is classified as "excellent".

[0051] When the grade value of the academic performance is between 0 and 100, the performance category is determined by the following thresholds: If the value is less than 60, it is classified as "unsatisfactory". If the value is between 60 and 69, inclusive, it is classified as "pass". If the value is between 70 and 79, inclusive, it is classified as "satisfactory". If the value is between 80 and 89, inclusive, it is classified as "good". If the value is 90 or above, it is classified as "excellent".

[0052] When the grade value of the academic performance is between 0 and 20, the performance category is determined by the following thresholds: If the value is less than 10, it is classified as "unsatisfactory". If the value is between 10 and 12, inclusive, it is classified as "pass". If the value is between 12 and 14, inclusive, it is classified as "satisfactory". If the value is between 14 and 16, inclusive, it is classified as "good". If the value is 16 or above, it is classified as "excellent".

[0053] In other embodiments, the numerical range of the score is set according to the actual situation. The score category can be set as a category with a score interval of 1, a score interval of 5, or a score interval of 20 minutes. The specific category setting is set by those skilled in the art, and the present invention does not make specific limitations on this.

[0054] Of course, the achievement category can be replaced with other categories, such as: dropout, graduation and further study. The present invention is not limited to this, and these equivalent technical solutions that can be obtained without creative work also belong to the protection scope of the present invention.

[0055] A lower predicted grade category may indicate that the student has issues managing their course load, understanding the material, or personal circumstances, which may affect academic achievement and continuity of learning.

[0056] S2. Based on the training data, a student score prediction model including a multi-layer CatBoost binary classifier is constructed, wherein the multi-layer CatBoost binary classifiers connected in sequence are respectively used to determine whether the student score belongs to each score category arranged from most to least.

[0057] Specifically, the distribution of students in different grades is unbalanced, for example, 5% of students have excellent grades, 25% of students have good grades, 40% of students have medium grades, 20% of students have passed grades, and 10% of students have failed grades. In the embodiment of the present invention, the students are sorted according to their distribution, and binary classification is performed each time to determine whether they belong to a grade category with the largest distribution of students; if not, continue to determine whether they belong to the next category until it is determined that the student to be predicted belongs to a certain category. The last layer of CatBoost binary classifier is used to determine which of the two categories with the least student distribution the academic performance of the student to be predicted belongs to. The problem of the model's prediction results being more biased towards the majority class is avoided. There will be no phenomenon in which the prediction results are more biased towards predicting that the student's grades are medium.

[0058] On the basis of the above embodiment, in an optional embodiment of the present invention, step S2 specifically includes steps S21 to S23.

[0059] S21. Obtain the number n of grades of the student's grades.

[0060] S22. According to the number of score categories, a student score prediction model is constructed by sequentially connecting n-1 CatBoost binary classifiers. The sequentially connected n-1 CatBoost binary classifiers are used to predict whether the student score belongs to the score category with the most students, whether it belongs to the score category with the second most students, whether it belongs to the score category with the third most students, and so on. The last CatBoost binary classifier is used to predict whether the student score belongs to the score category with the second least students.

[0061] S23. Among the n-1 CatBoost binary classifiers connected in sequence in the student score prediction model, when one CatBoost binary classifier predicts that the student score belongs to a certain score category, the prediction result is output; otherwise, the next layer of CatBoost binary classifiers continues to predict whether the student score belongs to the score category of the next student distribution, until the last layer of CatBoost binary classifiers completes the prediction and outputs the result of the last layer.

[0062] S3. According to the training data, cross-validation and grid search are used to train each layer of CatBoost binary classifiers of the student performance prediction model to obtain a trained student performance prediction model.

[0063] In an optional implementation, step S3 specifically includes step S31 to step S32.

[0064] S31. Based on the training data, 10-fold cross validation is performed: first, the data set is divided into 10 parts. In each iteration, one of the parts is used as a test set to verify the model performance, and the remaining 9 parts are used as training sets to train the model. This process is repeated 10 times, and the training set and test set are replaced each time. Finally, the average of the 10 results is used as the final performance evaluation of the model.

[0065] S32. Each time cross-validation is performed, the grade category with the most students in the training set is first divided into one category, and the remaining grade categories are divided into one category, and the first layer of CatBoost binary classifiers are trained. After completing the training of a CatBoost binary classifier, the training data of the grade category with the most students is deleted, and the grade category with the most students in the training data after the deletion is divided into one category, and the other grade categories are divided into one category, and the CatBoost binary classifier of the next layer is continued to be trained until the training of the last layer of CatBoost binary classifier is completed. Wherein, during training, the optimal binary classifier of the current layer is obtained based on grid search.

[0066] In an optional implementation, the optimal binary classifier of the current layer is obtained based on a grid search, specifically including: CatBoost binary classifier needs to adjust two hyperparameters, the number of base classifiers and the depth of each tree. Each hyperparameter has three candidate values, the candidate values ​​of the number of base classifiers are 100, 300 and 500, and the candidate values ​​of the depth of each tree are 3, 4 and 5. Among them, the candidate values ​​of the number of base classifiers and the candidate values ​​of the depth of each tree are randomly combined to obtain 9 pairs of hyperparameters [100,3], [100,4], [100,5], [300,3], [300,4], [300,5], [500,3], [500,4] and [500,5]. All these 9 pairs of hyperparameters are used for model training, and the best hyperparameter set is selected according to its performance.

[0067] Specifically, considering that the performance of the machine learning model is very sensitive to its hyperparameters, the embodiment of the present invention uses grid search and cross-validation to adjust the hyperparameters. Preferably, 10-fold cross-validation is used: first, the data set is divided into 10 parts, and in each iteration, one of them is used as a test set to verify the model performance, and the remaining 9 parts are used as training sets to train the model. This process is repeated 10 times, and the training set and test set are replaced each time, and finally the average of these 10 results is used as the final performance evaluation of the model.

[0068] In the training phase, the training set is first divided into two parts at the first layer: one part contains samples from the majority class (the output class corresponds to the data with the largest number of samples, for example, there is a data set of 100 samples, 40 students have excellent grades, 30 students have good grades, 15 students have medium grades, 10 students have passed grades, and 5 students failed, then the majority class samples are 40 excellent students), and the other part contains samples from the remaining classes (except for the samples of the most classes, that is, the remaining good, medium, passing, and failing classes are another class). This division helps to balance the two classes of samples. For the two divided classes of samples, the CatBoost binary classifier is trained using the training set, and the optimal binary classifier of the current layer is obtained based on grid search.

[0069] "Each pair of hyperparameters" refers to all possible combinations of the number of base classifiers and the depth of each tree. Specifically, these combinations include: ([100,3]): the number of base classifiers is 100, and the depth of each tree is 3. ([100,4]): the number of base classifiers is 100, and the depth of each tree is 4. ([100,5]): the number of base classifiers is 100, and the depth of each tree is 5. ([300,3]): the number of base classifiers is 300, and the depth of each tree is 3. ([300,4]): the number of base classifiers is 300, and the depth of each tree is 4. ([300,5]): the number of base classifiers is 300, and the depth of each tree is 5. ([500,3]): the number of base classifiers is 500, and the depth of each tree is 3. ([500,4]): The number of base classifiers is 500 and the depth of each tree is 4. ([500,5]): The number of base classifiers is 500 and the depth of each tree is 5.

[0070] Next, the majority class samples are deleted, and the remaining class samples are further divided into the majority class and the remaining class for the construction of the next layer of models. Sample deletion can reduce the number of training samples for the next layer of models, thereby improving the efficiency of model construction. Repeat the above steps until the remaining class contains only two classes, which are used to train the last layer of CatBoost model. At this time, the optimal CatBoost is trained and obtained, and the model construction is terminated to obtain a multi-layered student performance prediction model.

[0071] For example: If there is a data set of 100 samples, and the students' grades are excellent, good, medium, passing, and failing, 40 students have excellent grades, 30 students have good grades, 15 students have medium grades, 10 students have passed, and 5 students failed, these are the 5 categories.

[0072] When training the first layer of CatBoost binary classifier, the majority class is regarded as excellent and all other classes are regarded as another class.

[0073] When training the second layer of CatBoost binary classifiers, delete the majority class samples (i.e., data from the excellent class) when training the first layer, then set the good grade as the majority class, and set all other remaining classes as the other class. After training the second layer model for predicting the good class, delete the good class samples. Repeat this process to train the medium class prediction model and the passing class prediction model. Since there are only two classes left, the passing class and the failing class, the model building can be ended after training the current layer. When the predicted data does not belong to the passing class, it belongs to the last failing class with the least student distribution. At this time, the total number of layers is 4. In other words, assuming there are C given categories, the constructed model has C-1 layers, and each layer is a CatBoost binary classification model.

[0074] S4, obtaining the student data to be predicted, and inputting the trained student performance prediction model to obtain the student's predicted performance. Preferably, step S4 specifically includes steps S41 to S42.

[0075] S41. Obtain the student data to be predicted.

[0076] S42. Input the student data to be predicted into the trained student score prediction model, and use the multi-layer CatBoost binary classifier of the student score prediction model to predict layer by layer whether it belongs to the score category corresponding to the current layer. When it is predicted to belong to the score category corresponding to the current layer, the prediction result is output; otherwise, the data is input into the next layer for further judgment until the score category corresponding to the student data to be predicted is determined.

[0077] Specifically, for each sample in the test set / data set to be predicted, start from the first layer of the model and use exact matching. If the predicted category is consistent with the majority of categories in the current layer (for example, the first layer predicts that the excellent class model returns the result as excellent), then return to the current class and output the result; otherwise, enter the next layer for comparison (if the predicted result is another category, enter the second layer for comparison to see if the corresponding output result of the second layer is a good class. If it is a good class, return to the current category, otherwise enter the third layer). If the predicted category is consistent with the majority of categories in the current layer, then return to the current class and output the result. Repeat the above exact matching steps until the last layer, and output the final result.

[0078] The performance prediction device may be an electronic device with computing capabilities, such as a portable notebook computer, a desktop computer, a server, a smart phone, or a tablet computer.

[0079] It should be noted that most current student performance prediction models directly predict performance through multi-classification prediction, which involves collecting relevant data, cleaning and preprocessing, dividing training and test sets, and using training sets and grid search to optimize model parameters, and finally using test sets to evaluate model performance. However, due to the uneven distribution of student performance data categories, such as excellent only 5% and medium 40%, the model tends to predict the majority category, affecting the accuracy of the prediction. In the face of the challenge of class imbalance, commonly used sampling techniques such as oversampling and undersampling are used to deal with it. Oversampling alleviates the imbalance by increasing the number of minority class samples, but may introduce noise; while undersampling can reduce the number of majority class samples, it may lose important data and affect model performance.

[0080] In the embodiment of the present invention, considering that the multi-layer structure has stronger prediction ability than the single-layer model, CML-CatBoost (Class-Based Multi-Layer CatBoost) is used for student score prediction in multi-class classification tasks. In order to solve the class imbalance problem, a multi-layer structure is constructed to treat the majority class separately from other classes, effectively treating the problem as a binary classification task at each layer. First, the training set is divided into two parts: one part contains samples from the majority class and the other part contains samples from all other classes. This division helps to balance the representativeness of the two groups of samples, and then the binary classification model is trained on the first layer. Next, the samples belonging to the majority class are removed from consideration, and the remaining samples are retained to build the second layer. This process is repeated repeatedly to reduce the number of samples in a given category until only two categories are left in the data. In the prediction stage, an exact matching strategy is implemented layer by layer. This includes matching the predicted category with the target category in each layer. Once a match with the target category is found, the category is used as the final prediction result. Otherwise, it will enter the next layer until it is determined to match the target category.

[0081] The hierarchical grade prediction method of the present invention can effectively solve the problem of data category imbalance. The constructed model can capture and learn the complex patterns and nonlinear relationships behind student grades, thereby improving the accuracy of prediction.

[0082] The class-based multi-layer structure can process large-scale data sets more effectively. Each layer deletes the majority class samples, so that the model can still maintain an efficient computing speed when facing a large amount of student data. The design of the multi-layer structure helps the model generalize to different student groups and different educational environments, so that it is not limited to specific scenarios. Through the more accurate performance prediction of the model, it can help teachers and school administrators identify students who need additional help, so as to provide more personalized teaching support; it also provides data support for educational decision makers to help them adjust the allocation of teaching resources and educational strategies based on the prediction results. In addition, by predicting the performance trend of students, early intervention can be carried out before the students' academic performance declines, thereby improving the educational effect; through predictive analysis, educational resources can be allocated more reasonably, such as increasing teaching support for students with lower predicted performance. By predicting the possible difficulties and needs of students, teachers can adjust teaching methods and provide learning materials and activities that better meet the needs of students, thereby improving students' learning experience.

[0083] Embodiment 2: The present invention provides a hierarchical performance prediction device, which includes a training data acquisition module, a model construction module, a model training module and an inference module.

[0084] The training data acquisition module is used to acquire, clean and pre-process the training data. The training data includes multiple feature data of multiple students. The student scores are divided into multiple score categories. The model building module is used to build a student score prediction model including multiple layers of CatBoost binary classifiers according to the training data. The multiple layers of CatBoost binary classifiers connected in sequence are used to determine whether the student score belongs to each score category arranged from most to least students. The model training module is used to train each layer of CatBoost binary classifiers of the student score prediction model by cross-validation and grid search according to the training data to obtain a trained student score prediction model.

[0085] The inference module is used to obtain the student data to be predicted and input the trained student performance prediction model to obtain the student's predicted performance.

[0086] Embodiment 3: The present invention provides a hierarchical performance prediction device, comprising a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a hierarchical performance prediction method as described in any paragraph of the first aspect.

[0087] Embodiment 4: The present invention provides a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute a hierarchical performance prediction method as described in any paragraph of the first aspect.

[0088] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0089] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0090] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code. It should be noted that in this article, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such a process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0091] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0092] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0093] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0094] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A hierarchical performance prediction method, characterized in that: Include: Acquire training data, and perform cleaning and preprocessing; wherein the training data includes multiple feature data of multiple students; and the student scores are divided into multiple score categories; According to the training data, a student score prediction model including a multi-layer CatBoost binary classifier is constructed; wherein the multi-layer CatBoost binary classifiers connected in sequence are respectively used to determine whether the student score belongs to each score category arranged from most to least student distribution; Using cross validation and grid search according to the training data, each layer of CatBoost binary classifiers of the student score prediction model is trained to obtain a trained student score prediction model; The student data to be predicted is obtained, and the trained student performance prediction model is input to obtain the student's predicted performance.

2. A hierarchical performance prediction method according to claim 1, characterized in that: According to the training data, a student score prediction model including a multi-layer CatBoost binary classifier is constructed; wherein the multi-layer CatBoost binary classifiers connected in sequence are respectively used to determine whether the student score belongs to each score category arranged from most to least, specifically including: Get the number of grade categories n of the student's grades; According to the number of grade categories, a student grade prediction model is constructed in which n-1 CatBoost binary classifiers are connected in sequence; wherein the n-1 CatBoost binary classifiers connected in sequence are used to predict whether the student grade belongs to the grade category with the most students, whether it belongs to the grade category with the second most students, whether it belongs to the grade category with the third most students, and so on, and the last CatBoost binary classifier is used to predict whether the student grade belongs to the grade category with the second least students; Among the n-1 CatBoost binary classifiers connected in sequence in the student score prediction model, when one CatBoost binary classifier predicts that the student score belongs to a certain score category, it outputs the prediction result, otherwise the next layer of CatBoost binary classifier continues to predict whether the student score belongs to the score category of the next student distribution, until the last layer of CatBoost binary classifier completes the prediction and outputs the result of the last layer.

3. A hierarchical performance prediction method according to claim 1, characterized in that: Using cross validation and grid search according to the training data, each layer of CatBoost binary classifiers of the student score prediction model is trained respectively to obtain the trained student score prediction model, which specifically includes: Based on the training data, 10-fold cross validation was used: first, the data set was divided into 10 parts. In each iteration, one of the parts was used as a test set to verify the model performance, and the remaining 9 parts were used as training sets to train the model. This process was repeated 10 times, and the training set and test set were replaced each time. Finally, the average of the 10 results was used as the final performance evaluation of the model. In each cross-validation, the grade category with the most students in the training set is first divided into one category, and the remaining grade categories are divided into one category, and the first layer of CatBoost binary classifiers are trained; after completing the training of a CatBoost binary classifier, the training data of the grade category with the most students is deleted, and the grade category with the most students in the training data after the data is deleted is divided into one category, and the other grade categories are divided into one category, and the next layer of CatBoost binary classifiers are trained until the training of the last layer of CatBoost binary classifiers is completed; during training, the optimal binary classifier of the current layer is obtained based on grid search.

4. A hierarchical performance prediction method according to claim 3, characterized in that: The optimal binary classifier of the current layer is obtained based on grid search, including: The CatBoost binary classifier needs to adjust two hyperparameters, the number of base classifiers and the depth of each tree. Each hyperparameter has three candidate values, the candidate values ​​of the number of base classifiers are 100, 300 and 500, and the candidate values ​​of the depth of each tree are 3, 4 and 5. Among them, the candidate values ​​of the number of base classifiers and the candidate values ​​of the depth of each tree are randomly combined to obtain 9 pairs of hyperparameters [100,3], [100,4], [100,5], [300,3], [300,4], [300,5], [500,3], [500,4] and [500,5]. All these 9 pairs of hyperparameters are used for model training, and the best hyperparameter set is selected according to its performance.

5. A hierarchical performance prediction method according to claim 1, characterized in that: Obtain the student data to be predicted, and input the trained student performance prediction model to obtain the student's predicted performance, specifically including: Get the student data to be predicted; The student data to be predicted is input into the trained student score prediction model, and the multi-layer CatBoost binary classifier of the student score prediction model predicts layer by layer whether it belongs to the score category corresponding to the current layer. When it is predicted to belong to the score category corresponding to the current layer, the prediction result is output; otherwise, the data is input into the next layer for further judgment until the score category corresponding to the student data to be predicted is determined.

6. A hierarchical performance prediction method according to claim 1, characterized in that: The training data is obtained, cleaned and preprocessed; wherein the training data includes multiple feature data of multiple students; the student scores are divided into multiple score categories, specifically including: Acquire training data; wherein the training data comprises multiple feature data of multiple students; the features of the feature data comprise academic performance, demographic information, social background data and school-specific features; wherein the academic performance comprises the school the student attends, the weekly study time, the number of past course failures, the first semester grades, the second semester grades, and the final grades; the demographic information comprises the student's gender, the student's age, the student's nationality, and whether the student is an international student; the social background data comprises free time after school, extracurricular activities, the type of student's home address, the father's education level, the mother's education level, the father's occupation, the mother's occupation, and the time from home to school; the school-specific features comprise the reason for choosing this school, the number of absences from school, whether the student has any special educational needs, the number of course units approved by the student in the first semester, the number of course units approved by the student in the second semester, the average grade in the first semester, and the average grade in the second semester; The training data is cleaned and preprocessed, and the feature importance of each feature to the learning performance is obtained through a random forest model with default hyperparameters; Filter out useful features according to the feature importance and obtain useful training data; According to the useful training data, the student scores are divided into a plurality of score categories to obtain a final training data set.

7. A hierarchical performance prediction method according to claim 6, characterized in that: When the grade value of the learning performance is between 0 and 100, the performance category is determined by the following thresholds: If the value is less than 60, it is classified as "failed"; If the value is between 60 and 69, inclusive, it is classified as "qualified"; If the value is between 70 and 79, inclusive, it is classified as "satisfactory"; If the value is between 80 and 89, inclusive, it is classified as "good"; If the value is 90 or above, it is classified as "excellent".

8. A hierarchical performance prediction device, characterized in that: Include: The training data acquisition module is used to acquire training data and perform cleaning and preprocessing; wherein the training data includes multiple feature data of multiple students; the student scores are divided into multiple score categories; A model building module is used to build a student score prediction model including a multi-layer CatBoost binary classifier according to the training data; wherein the multi-layer CatBoost binary classifiers connected in sequence are respectively used to determine whether the student score belongs to each score category arranged from most to least student distribution; A model training module, used to train each layer of CatBoost binary classifiers of the student performance prediction model by cross-validation and grid search according to the training data, to obtain a trained student performance prediction model; The inference module is used to obtain the student data to be predicted and input the trained student performance prediction model to obtain the student's predicted performance.

9. A hierarchical performance prediction device, characterized in that: It comprises a processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement a hierarchical performance prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute a hierarchical performance prediction method as described in any one of claims 1 to 7.

Images