Academic performance prediction method and accademic performance prediction system

Abstract

The invention provides an academic performance prediction method, which comprises the steps of collecting first data for academic performance, and preprocessing the first data to obtain second data, the preprocessing comprising data filtering and data coding of the first data; converting the second data into input features, inputting the input features into the target model, and generating prediction probability distribution corresponding to the academic performance; and carrying out weighted fusion on the prediction probability distribution, generating a prediction result corresponding to the academic performance, and carrying out associated storage on the prediction result and the student identification information. The limitation that academic performance evaluation is carried out only by means of a single cognitive factor is overcome, the overall data of family background, psychological health and academic performance corresponding to each student is collected, family background data and psychological evaluation data serve as non-cognitive factors, the non-cognitive factors are integrated, processed and predicted through a deep learning framework, and the accuracy of academic performance evaluation is improved. And deep understanding and accurate prediction of academic performance of students are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data analysis, in particular to a method and system for predicting academic performance. BACKGROUND

[0002] The average grade point average (GPA) is also known as academic performance, which is the average grade point average calculated based on the scores of each course during a certain learning period. Generally, academic performance is determined by cognitive factors, including intelligence, knowledge, learning attitude, academic potential, etc.

[0003] With the development of society, non-cognitive factors have an increasingly important impact on academic performance. Non-cognitive factors refer to a series of psychological and social variables that affect learning and behavior performance, including motivation, emotion, mental health, interpersonal relationships, family environment, etc. Non-cognitive factors cannot directly measure academic performance, but they deeply affect the learning process and continuous investment of students. Therefore, a method is needed to predict academic performance based on non-cognitive factors to achieve intelligent teaching support and personalized intervention. SUMMARY

[0004] The present application provides a method for predicting academic performance to solve the problem of inaccurate academic performance determination by cognitive factors.

[0005] The present application provides a method for predicting academic performance, comprising: collecting first data related to academic performance, preprocessing the first data to obtain second data, wherein the preprocessing includes data filtering and data encoding of the first data; converting the second data into input features, inputting the target model, and generating a prediction probability distribution corresponding to the academic performance; weighting and fusing the prediction probability distribution to generate a prediction result corresponding to the academic performance, and storing the prediction result in association with the student identification information.

[0006] In an embodiment of the present application, the first data includes at least one of family background data, psychological assessment data and academic performance data, wherein the data form of the psychological assessment data includes a symptom self-assessment scale, and the symptom self-assessment scale includes factors and item scores.

[0007] In an embodiment of the present application, preprocessing the first data to obtain the second data includes: removing psychological assessment data with an answer time lower than a preset answer time.

[0008] In an embodiment of the present application, the method further comprises: determining a data type corresponding to the first data; for the data type being numerical data, encoding the numerical value of the first data; for the data type being categorical data, encoding by using integer category mapping, wherein the integer category mapping comprises at least one of 0 / 1 encoding and continuous numerical encoding.

[0009] In an embodiment of the present application, the method of converting the second data into input features, inputting the target model, and generating the prediction probability distribution corresponding to the academic performance comprises: mapping the numerical data included in the second data to a preset numerical range to obtain numerical features; performing embedding operation on the categorical data included in the second data to obtain categorical features; inputting the numerical features, the categorical features, and the merged features into at least one target model to generate the prediction probability distribution.

[0010] In an embodiment of the present application, the target model comprises at least one of sub-models for processing numerical features, categorical features, and merged features, and the weight of the sub-model is optimized and adjusted when the sub-model processes the numerical features, the categorical features, and the merged features. In an embodiment of the present application, the method further comprises: evaluating the sub-models so that the weights of the sub-models corresponding to the same numerical features, categorical features, and merged features are the same; and performing correlation analysis on the sub-models to adjust the analysis-prediction ability of the sub-models for the numerical features, the categorical features, and the merged features.

[0011] In an embodiment of the present application, the method of weighting and fusing the prediction probability distribution to generate the prediction result corresponding to the academic performance, and storing the prediction result in association with the student identification information comprises: receiving the prediction probability distribution from at least one sub-model; fusing and calculating the prediction probability distribution of each sub-model according to the fusion weight corresponding to the sub-model to obtain a weighted average probability as the prediction result; and storing the prediction result and the student identification information in the form of key-value pairs.

[0012] The second aspect of the present application discloses an academic performance prediction system, comprising: a data acquisition module, a data preprocessing module, a training and testing module, and a prediction module. The data acquisition module is used to acquire first data corresponding to academic performance, the data preprocessing module is used to preprocess the first data to obtain second data, wherein the preprocessing comprises data filtering and data encoding of the first data; the training and testing module is used to convert the second data into input features, input at least one target model, and generate a prediction probability distribution corresponding to the academic performance; and the prediction module is used to weight and fuse the prediction probability distribution to generate a prediction result corresponding to the academic performance, and store the prediction result in association with student identification information.

[0013] The third aspect of the present application discloses an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the academic performance prediction method according to the first aspect when executing the computer program The present application has the beneficial effects that: through the academic performance prediction method, the limitation of academic performance evaluation by only a single cognitive factor is overcome, the overall data of the corresponding family background, mental health and academic performance of each student is collected, wherein the family background data and the mental evaluation data are non-cognitive factors, the non-cognitive factors are integrated, processed and predicted through a deep learning framework, the deep understanding and accurate prediction of the academic performance of the student are realized, and a technical solution with accurate prediction is formed. BRIEF DESCRIPTION OF DRAWINGS

[0014] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and serve to explain the principles of the present application. It is readily apparent to one skilled in the art that the following description in the drawings is merely some embodiments of the present application, and other drawings can be obtained from these drawings without creative labor.

[0015] In the drawings: Figure 1 a flowchart of the academic performance prediction method provided in an embodiment of the present application; Figure 2 a schematic diagram of the academic performance prediction system provided in an embodiment of the present application; Figure 3 a structural schematic diagram of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The embodiments of the present application will be described in detail below with specific reference to the drawings. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. The present application can also be implemented or applied in different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the present application. In the case of no conflict, the following embodiments and features in the embodiments can be combined with each other.

[0017] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present application in a schematic manner, and the drawings only show the components related to the present application, not the number, shape and size of the components when actually implemented. The actual implementation of each component may be a random change, and the component layout pattern may be more complex.

[0018] In the following description, numerous specific details are discussed in order to provide a thorough explanation of embodiments of the application. However, it will be apparent to one of ordinary skill in the art that embodiments of the application can be practiced without these specific details. In other instances, well-known structures and devices are not described in detail in order to avoid obscuring embodiments of the application.

[0019] In some embodiments, the non-cognitive factors involved in embodiments of the application can include family background data, psychological assessment data, and the like. Among them, the family background data can include gender, whether only child, family economic status, stay time, and the like; the psychological assessment data can be determined based on a symptom checklist-90 (SCL-90), which mainly includes psychopathology, involving data such as feeling, thinking, emotion, behavior, interpersonal relationship, living habit, diet, sleep, and the like, and reference Table 1, which shows an example of a symptom checklist-90. As can be seen, the psychological assessment data can include factor and item score, for example, factor: headache, item score: none, mild, moderate, severe, and the like. Compared with the non-cognitive factors, the cognitive factors can be directly represented by academic performance data.

[0020]

[0021] Table 1 Based on the above non-cognitive factors, embodiments of the application propose a method for predicting academic performance, including: a data collection step of collecting first data (academic performance evaluation data) for academic performance, wherein the first data can include at least one of family background data, psychological assessment data, and academic performance data, and can also be referred to as multi-dimensional data; a data preprocessing step of preprocessing the first data to obtain second data, the preprocessing including: data selection, data filtering, and data encoding on the first data, the data selection and data filtering being used to remove noise data, and the data encoding being used to encode the first data according to at least one data type corresponding to the first data, the data type including categorical data and numerical data. A multi-model prediction step of converting the second data into input features, wherein the input features correspond to the data types, inputting at least one target model, and generating a prediction probability distribution corresponding to the academic performance. A result fusion and storage step of weighting and fusing the prediction probability distributions output by the multiple deep learning models to generate a prediction result corresponding to the academic performance, and associating and storing the prediction result with student identification information.

[0022] It can be seen that by the above-mentioned academic performance prediction method, the limitation of evaluating academic performance by only relying on a single cognitive factor is overcome, the overall data of the corresponding family background, mental health and academic performance of each student is collected, among which the family background data and the psychological assessment data are non-cognitive factors, the non-cognitive factors are integrated, processed and predicted through a deep learning framework, the deep understanding and accurate prediction of the academic performance of the student are realized, and a prediction accurate technical scheme is formed, which has important practical significance and social value for improving the quality of education, promoting education equity and protecting the healthy development of students.

[0023] Please refer to Figure 1 , Figure 1 The academic performance prediction method provided by an embodiment of the present application, as shown in Figure 1 , the prediction method comprises the following steps.

[0024] Step S101: collecting first data for academic performance, wherein the first data.

[0025] In some embodiments, the first data herein can also be referred to as academic performance evaluation data, including multi-dimensional student data, for example: student data can cover family background data, psychological assessment data and academic performance data, which can ensure the comprehensiveness, objectivity and comparability of student data, and the first data can also be referred to as multi-dimensional data.

[0026] The family background data can include individual basic demographic information, family socio-economic status indicators, family structure and peer support information, and family guardianship and accompanying conditions, wherein the individual basic demographic information includes, for example: gender: record the physiological gender of the student, and when analyzing academic performance or psychological symptoms, male and female can be used as independent control groups. Age: record the age of the student in years, which is used to control the influence of the development stage. Grade: record the current grade of the student (such as junior two, senior one), which is used to distinguish different education stages and academic requirements.

[0027] The family socio-economic status indicators are used to quantify the material and cultural resources that the family can provide, including, for example: residential type: divided according to urban and rural attributes, for example: divided into “city”, “county town”, “township”, “rural” four levels. The residential type can reflect the regional development level and the accessibility of educational resources. Family economic status: measured by family monthly or annual income level, for example: divided into “poor”, “general”, “rich” two subcategories. Parental education level: collect the highest education level of both parents, usually divided into “primary school and below”, “junior high school”, “high school / technical school”, “college / university”, “master and above” and the like.

[0028] Information on family structure and peer support includes, for example: Family structure: The core indicator is "whether the child is an only child." For example, mark "yes" or "no." Peer support: Assess the student's perceived level of peer support through a simplified scale or key questions. For example, ask, "When you encounter difficulties, do you have any close friends who can help you?" and categorize them as "almost none," "occasionally," "often," "always," etc.

[0029] Family guardianship and companionship include, for example: Left-behind experience: Determining whether a student has experienced being left behind, i.e., both parents or one parent being away for work, business, or study for more than 6 consecutive months, categorized as "yes" or "no". Duration of being left behind: If left-behind experience exists, its cumulative duration needs to be collected. It is preferable to record it as an integer value in "years," typically covering 0 to 18 years (i.e., the entire period of childhood). For example, if a student was under the guardianship of their grandparents during primary school (ages 6-12), and both parents worked in a coastal city, then their "duration of being left behind" would be recorded as 6 years. Type of being left behind: Further distinguishing between different types such as "none," "father away," "mother away," or "both parents away," to explore the differences in the impact of different guardianship absence patterns.

[0030] The data format of psychological assessment data can include symptom self-rating scales. Based on the factors and item scores selected by the student on the symptom self-rating scale, multiple factor scores and a total score for the student's corresponding symptom self-rating scale can be calculated.

[0031] Academic performance data refers to a student's GPA, such as 3.45 (out of 4.0). Academic performance is a standardized indicator that comprehensively reflects a student's learning level in various courses, and it is comprehensive and comparable.

[0032] As can be seen, family background data can be obtained through parent questionnaires or student basic information forms; psychological assessment data can be obtained through online or offline questionnaires using symptom self-rating scales; and academic performance data can be obtained through the school's academic affairs management system. The above data form the initial dataset for subsequent analysis.

[0033] Step S102: Preprocess the first data to obtain the second data.

[0034] In some embodiments, the data preprocessing herein can include data selection, data filtering and data encoding. Before using the data for training and prediction, the data needs to be selected and filtered to ensure the quality and representativeness of the data. For example: the distribution of the data should be as balanced as possible to avoid the model from being biased due to too much data of a certain type. For example, in terms of gender distribution, it can be ensured that the male and female ratio corresponding to the data is close to 1:1. The data should have diversity in dimensions such as grade, type of residence, family economic status, so that the model trained based on the data has good generalization ability. After the above data selection processing, the selected data also needs to be filtered and encoded, and the embodiments of the present application can adopt the quality control filtering and standardized encoding method.

[0035] Specifically, for the psychological assessment data collected by the symptom self-rating scale, in order to avoid introducing invalid symptom self-rating scales into multi-dimensional data to form noise data, quality control filtering can include response time filtering and response mode filtering. Response time filtering (collection time filtering): the symptom self-rating scale contains approximately 90 items, and serious response usually requires a certain amount of time. Symptom self-rating scales with response time less than a preset response time (e.g., 200 seconds) can be removed. For example, a student completes the filling of all 90 questions in the symptom self-rating scale in only 150 seconds, which is not in line with the required time for normally responding to the symptom self-rating scale, and is judged to be superficial and is removed.

[0036] Response mode filtering: In addition to the total response time, abnormal response mode is also a sign of invalid response. Data with the same option selected for 30 consecutive options in the symptom self-rating scale (e.g., all selecting "none" or all selecting "severe") can be removed. For example, a student selects "2. Mild" without hesitation for 30 consecutive questions from question 31 to question 60 of the SCL-90 scale, which does not conform to the reporting rules of human real and fluctuating psychological feelings and is considered as an invalid response mode.

[0037] As can be seen, through the above data selection and data filtering, the authenticity and effectiveness of the data used for training and prediction can be maximized.

[0038] Data encoding is used to uniformly convert different data types into data suitable for machine learning model processing, including discretization or normalization of data, etc. The data performance after data encoding is different for different data types. For example, for numerical data, direct encoding can be used, that is, directly using numerical values. For example, for GPA, the value range of GPA is [0, 4], and the GPA encoding method can directly use the original numerical value. For example, a student's GPA directly uses 3.45.

[0039] The encoding method of the category type data is that each category of the category type data is mapped to a unique integer. For example, for the data of family economic status, the original data can be "poor", "general" and "rich", and the present application assigns values of 0, 1 and 2 to them respectively, that is, continuous numerical encoding. This encoding method can convert the category type data (text information) into numerical values, which is convenient for model processing.

[0040] The encoding method of the category type data also includes the encoding method of binary classification data (0 / 1). For binary classification data with only two states, 0 and 1 are used for encoding, that is, 0 / 1 encoding. For example, for gender: "male" is encoded as 0 and "female" is encoded as 1. For whether the child is an only child: "yes" is encoded as 1 and "no" is encoded as 0.

[0041] The process of step S102 is illustrated below. The data determined through step S101 can include: data 1 and data 2, wherein data 1: student number: S001, gender: female, age: 16, type of residence: rural, family economic status: poor, whether an only child: yes, duration of stay: 6, type of stay: both parents out, psychological assessment data: 2.8, GPA: 2.89. Data 2: student number: S002, gender: male, age: 17, type of residence: city, family economic status: general, whether an only child: yes, duration of stay: 0, type of stay: none, psychological assessment data: 1.8, GPA: 3.75. After data preprocessing through step S102, data 1: student number: S001, gender: 1, age: 16, type of residence: 0, family economic status: 0, whether an only child: 1, duration of stay: 6, type of stay: 1, psychological assessment data: 2.8, GPA: 2.89. Data 2: student number: S002, gender: 0, age: 17, type of residence: 1, family economic status: 1, whether an only child: 0, duration of stay: 0, type of stay: 0, psychological assessment data: 1.8, GPA: 3.75. In data 1 and data 2, gender, type of residence, whether an only child and type of stay are category type data.

[0042] It can be seen that through data selection and data filtering, unreliable data can be removed, noise can be significantly reduced, validity and reliability of subsequent training and prediction can be improved, different types of data are standardized and encoded, the uniformity of data format is ensured, and input data that can be directly calculated is provided for the academic performance prediction model, so that the academic performance prediction model can further extract features from the data.

[0043] Step S103: inputting the second data into the academic performance prediction model for training and prediction.

[0044] In some embodiments, since the second data (input data) preprocessed through step S102 includes at least two types of data, numerical data and categorical data, when learning and predicting the target data, a model good at processing the features corresponding to the numerical data and the categorical data can be used, so that the prediction result is more accurate. The academic performance prediction model herein is used to predict the GPA level of a student, and can include at least four sub-models, such as a TabTransformer model, a DCNv2 model, an AutoInt model and an MLP-ResNet model. Different models can process the input data at the same time, but different learning strategies are adopted for the input data. For example, the TabTransformer model is good at processing the categorical data in the input data. The TabTransformer model first converts each categorical data into a dense vector representation (embedding vector), and then allows different features to pay attention to and communicate with each other through the self-attention mechanism of the Transformer. For example, when predicting the GPA, the model can automatically pay attention to the "residence type" feature and the "family economic status" feature, and learn the deep relationship between them. The input format of the TabTransformer model includes categorical features and numerical features, which are input separately, and the TabTransformer model performs embedding operation on the categorical features, and generates output features through average pooling of the categorical features and the numerical features.

[0045] The DCNv2 model can implicitly learn the complex high-order interaction between features through a deep neural network (DNN), and explicitly cross the features through a special cross network to explicitly calculate the cross product relationship between the features. The input format of the DCNv2 model includes merged categorical features and numerical features, and the DCNv2 model performs embedding operation on the categorical features and merges the numerical features to obtain merged features.

[0046] The AutoInt model focuses on mining high-order interactions between features through a multi-head attention mechanism. It regards all features (including categorical and numerical features) as a sequence, and allows each feature to interact with all other features through an attention mechanism. The multi-head design allows the model to focus on different types of relationship patterns at the same time. Residual connection ensures that information can be effectively transmitted in the deep network, preventing gradient disappearance. The input format of the DCNv2 model includes categorical features and numerical features that have been subjected to embedding operation, and both of them need to be subjected to embedding operation.

[0047] The MLP-ResNet model, as a deep nonlinear mapper, adopts a residual connection structure similar to ResNet. Each layer of the network retains a shortcut connection of the original input when calculating a new feature representation, which enables stable gradient propagation even in very deep networks. The input format of the MLP-ResNet model includes merged categorical features and numerical features.

[0048] Before processing the input data through the academic performance prediction model, the second data (input data) needs to be processed to obtain input features, including feature standardization operation, embedding operation, and merging operation. The following takes data 1 in step S102 as an example to explain in detail how to perform feature standardization operation, embedding operation, and merging operation on numerical data and categorical data, etc. The values of the categorical data included in data 1 can be [1, 0, 0, 1, 1], i.e., gender: female (1) -> embedding vector [0.1, -0.2, 0.3, 0.4], residence: rural (0) -> embedding vector [-0.5, 0.6, 0.7, -0.8], family economic status: poor (0) -> embedding vector [0.9, 1.0, -1.1, 1.2], only child: yes (1) -> embedding vector [-1.3, 1.4, 0.5, -0.6], and stay type: both parents out (1) -> embedding vector [0.7, -0.8, 0.9, 1.0]. After performing the concatenation operation, the categorical features are [0.1, -0.2, 0.3, 0.4, -0.5, 0.6, 0.7, -0.8, -1.3, 1.4, 0.5, -0.6, 0.7, -0.8, 0.9, 1.0]. The numerical data can be [16, 6, 2.8, 2.89], and the numerical data is subjected to feature standardization operation (after, the numerical features are obtained: age: -1.0, stay duration: 0.71, psychological assessment data: 0.71, GPA: -0.71. The feature standardization operation here includes mapping the numerical data to a predetermined numerical range, such as the [0, 1] range. After performing the merging operation on the numerical features and the categorical features, the input features are [0.1, -0.2, 0.3, 0.4, -0.5, 0.6, 0.7, -0.8, 0.9, 1.0, -1.1, 1.2, -1.3, 1.4, 0.5, -0.6, 0.7, -0.8, 0.9, 1.0, 0.71, 0.71, -0.71]. As can be seen, the input features are a 24-dimensional vector, and as the dimension of the student data increases, the dimension of the input features will also increase.

[0049] It can be understood that before the training starts, the original table data is processed. The numerical features are standardized or normalized to eliminate the dimensional difference; the category features are encoded into a format suitable for neural network processing. The input data after feature processing can also be divided into training set, validation set and test set to ensure the objectivity of the academic performance prediction model. In the training stage, consistent data division, hyperparameter grid / Bayesian search and early stopping strategy configuration can be performed on each sub-model. In the prediction stage, the main indicators and auxiliary indicators are reported under the same evaluation protocol, and statistical significance test and ablation analysis are performed to ensure the fairness and reproducibility of the comparison between different benchmarks.

[0050] Step S104: Perform fusion feature gating processing and prediction on the academic performance prediction model.

[0051] In some embodiments, since the academic performance prediction model needs to process a large number of high-dimensional input features, the requirements for feature screening and robustness are high. Therefore, a lightweight feature gating mechanism (Gating module, also known as gating module) can be selected as a pre-feature weighting layer of the academic performance prediction model, which can dynamically adjust the contribution of different sub-models to the prediction result, realize adaptive fusion at the feature level, that is, optimize and adjust the weights of each sub-model. The gating module, as a lightweight neural network component, is embedded in each sub-model to optimize and adjust the weights.

[0052] For the TabTransformer model, a gating module is added after the Transformer encoder, that is, the category features and numerical features are concatenated and weighted by the gating module. After each feature is encoded by the Transformer encoder, the importance weight is generated by the gating module, and then multiplied element-wise with the original feature: h'_i=G_i(h_i)⊙h_i, where h_i is the representation of the i-th feature after Transformer encoding, G_i is the gating function learned for the feature.

[0053] For the DCNv2 model, independent gating is added to the output of the implicit DNN path and the explicit cross network path, and the weighted sum of the two path outputs is performed: output=α·G_dnn(DNN(x))+(1-α)·G_cross(CrossNet(x)), where α is learned from data, and G_dnn and G_cross correspond to the gating functions of the two paths, respectively.

[0054] For the AutoInt model, a head-level gating is added to each head of the multi-head attention mechanism, and a layer-level gating is added between layers, with a series structure as the weighted layer of the input end. The head-level gating learns the importance of different attention patterns, and the layer-level gating adjusts the contribution of different depths: Head_i'=β_i·Attention_Head_i(Q, K, V), Layer_j'=γ_j·∑(Head_i').

[0055] For the MLP-ResNet model, a path gating is added between the shortcut connection and the main path of the residual block. This allows the model to dynamically decide to retain the original information and learn new features: output=λ·F(x)+(1-λ)·x, where λ is the gating parameter and F(x) is the main path transformation of the residual block.

[0056] It can be seen that the practical value of the gating module in the table data deep learning model provides data-driven guidance for model architecture design.

[0057] Step S105: Model evaluation is performed on the academic performance prediction model.

[0058] In some embodiments, in the academic performance prediction model, the above four models can also be subjected to model evaluation and correlation analysis for verifying the effectiveness of the academic performance prediction model.

[0059] For example: the model evaluation of the academic performance prediction model can use various classification indicators, including Accuracy, Macro-Precision, Macro-Recall, Macro-F1Score and AUC, etc., to comprehensively measure the performance of the model in predicting the GPA level prediction. These classification indicators are calculated based on the test set corresponding to the input data, ensuring that each GPA level (excellent, good, and medium) is given equal weight. Specifically, the evaluation process first loads the weights of the trained sub-models, and then generates the prediction probability distribution on the test set.

[0060] Among them, Accuracy represents the proportion of samples whose output prediction results are correct in the total number of samples in the test set, which is a direct indicator for measuring the academic performance prediction model. Macro-Precision is the proportion of actual positive samples in the samples predicted by the model as positive. Macro-Recall is the proportion of actual positive samples correctly predicted by the model. Macro-F1 is the harmonic mean of macro-precision and macro-recall, and F1 score can comprehensively reflect the balance of the model in precision and recall. In a multi-class classification task, a one-to-one strategy is adopted, AUC is calculated for each pair of classes, and then the AUC of all class pairs is averaged to comprehensively evaluate the discriminant ability of the model at different thresholds.

[0061]

[0062]

[0063]

[0064]

[0065]

[0066] wherein, K represents the number of categories, TPk represents the positive samples of the Kth category, N represents the total number of samples, FPk represents the false positive of the Kth category, FNk represents the false negative of the Kth category, and AUCi,j represents the AUC value between the ith category and the jth category.

[0067] It can be seen that by using multiple classification indicators to evaluate the prediction accuracy and generalization ability of each sub-model in the academic performance prediction model, it is ensured that the academic performance prediction model can realize high-precision prediction.

[0068] Step S106: Correlation analysis on the academic performance prediction model In some embodiments, the correlation is used to represent the complex relationship between the family background data, psychological assessment data and academic performance data in the input data, and one or more methods including Spearman rank correlation, chi-square test and t-SNE visualization can be used. Among them, the Spearman rank correlation is used to calculate the monotonic correlation coefficient r and the p value between the features, the chi-square test is used to test the independence between the categorical features; t-SNE as a nonlinear dimensionality reduction technique, first applies PCA pre-dimensionality reduction, then maps high-dimensional data to two-dimensional space, and displays sample clustering in scatter plot to reveal natural distribution patterns and potential groups; the analysis process includes data preparation, method application and visualization generation, first calculate the Spearman correlation pairwise to form a correlation matrix heat map, then perform chi-square test on all categorical features, and finally t-SNE projection reveals the division of the feature pair to the sample space. This analysis not only quantifies the correlation strength, but also provides visual evidence to support the practicability of the framework in identifying academic risks.

[0069] Specifically, the correlation analysis (Spearman correlation, Chi-square test, t-SNE) and each model included in each sub-model in the academic performance prediction model can realize the two-way enhancement of explanatory analysis-predictive modeling. The correlation analysis of the TabTransformer model sets a higher initial attention preference for high-correlation feature pairs such as depression-anxiety. The correlation analysis of the DCNv2 model prioritizes the inclusion of cross-networks for feature pairs with high correlation coefficients. The correlation analysis of the AutoInt model designs information interaction between attention heads. The correlation analysis of the MLP-ResNet model sets more identity connections for strongly correlated features.

[0070] Step S107: output the prediction result.

[0071] Model parallel inference and probability generation.

[0072] In some embodiments, taking the data of student ID "S2023001" as an example, the academic performance prediction system receives a request for academic risk prediction for student "S2023001". After the input features of the student are input into the academic performance prediction model, each model inside the model deeply mines the complex relationships between features with its unique architecture (such as self-attention of TabTransformer, cross-network of DCNv2, multi-head interaction of AutoInt, and deep residual mapping of MLP-ResNet), and finally produces an original output score at its output layer. A Softmax activation function can be connected after the output layer of each model. This function converts the original output score of the model into a prediction probability distribution. This distribution contains three probability values corresponding to the encoding mode of "excellent", "good", and "average" GPA, respectively, and each probability value is between 0 and 1, and the sum of the three is 1.

[0073] For example: for student "S2023001", the four models can output the following prediction probability distributions (example values): TabTransformer model output: P(excellent)=0.10, P(good)=0.75, P(average)=0.15, DCNv2 model output: P(excellent)=0.05, P(good)=0.80, P(average)=0.15, AutoInt model output: P(excellent)=0.08, P(good)=0.70, P(average)=0.22, MLP-ResNet model output: P(excellent)=0.12, P(good)=0.68, P(average)=0.20.

[0074] Model result fusion and final prediction probability distribution calculation In some embodiments, the academic performance prediction model can be provided with a fusion module (e.g., a weighted average module) to receive the prediction probability distribution from the four models. The fusion module can assign a fusion weight to each model. For example, the four models are assigned weights of 0.25, 0.30, 0.25, and 0.20, respectively. The fusion module calculates the weighted average probability of the prediction probability distribution corresponding to each GPA level.

[0075] For example, the prediction result corresponding to the "good" GPA level for student "S2023001" can be represented as P_final(good) = 0.25 * 0.75 + 0.30 * 0.80 + 0.25 * 0.70 + 0.20 * 0.68 = 0.734.

[0076] Similarly, the probabilities of other levels are calculated to obtain the fused prediction probability distribution corresponding to student S2023001 as the prediction result, for example: prediction probability distribution: P(excellent) = 0.087, P(good) = 0.734, P(average) = 0.179 Key-value pair structured storage of prediction results In some embodiments, the academic performance prediction system is designed to ensure data consistency, traceability, and support for possible subsequent analysis. The generated prediction results are stored in a structured manner in a storage area (e.g., a database). The storage area can be stored in the form of key-value pairs (Key / ValueP), where the key (Key) is the student ID, i.e., "S2023001", and the value (Value) is the prediction result, for example: {"excellent": 0.087, "good": 0.734, "average": 0.179}.

[0077] As can be seen, the prediction results corresponding to the GPA levels are temporarily stored in the form of key-value pairs, so that the prediction results of each student can be preserved, facilitating problem troubleshooting and model effectiveness analysis. The prediction results contain the final probability distribution of each GPA level, allowing users of the academic performance prediction model to assess the degree of confidence in the prediction. At the same time, the formatted GPA level prediction results are easy to integrate and analyze by other systems (e.g., student management systems, early warning platforms).

[0078] Please refer to Figure 2 , Figure 2This is a schematic diagram of the modules of an academic performance prediction system provided in an embodiment of the present invention. The academic performance prediction system 20 may include: a data acquisition module 201, a data preprocessing module 202, a training and testing module 203, and a prediction module 204. The data acquisition module 201 is used to collect first data related to academic performance; the data preprocessing module 202 is used to preprocess the first data to obtain second data; the training and testing module 203 is used to convert the second data into input features, input them into at least one target model, and generate a prediction probability distribution corresponding to academic performance; the prediction module 204 is used to perform weighted fusion of the prediction probability distribution to generate a prediction result corresponding to academic performance, and associate and store the prediction result with student identification information. The training and testing module 203 further includes: a weight optimization submodule 2031, an evaluation submodule 2032, and a correlation analysis submodule 2033. The weight optimization submodule 2031 is used to set a pre-feature weighting layer (i.e., a gating module) for each sub-model in the academic performance prediction model, which can dynamically adjust the contribution of different sub-models to the prediction result to achieve adaptive fusion. The gating module, a lightweight network component, is embedded in each sub-model and is responsible for learning the weights of the input features. The evaluation sub-module 2032 employs various classification metrics, including Accuracy, Macro-Precision, Macro-Recall, Macro-F1Score, and AUC, to measure the performance of each sub-model in predicting GPA grades. These metrics are calculated based on the test set corresponding to the input data, ensuring that each GPA grade (Excellent, Good, Average) receives equal weight. The correlation analysis sub-module 2033 processes each sub-model in the academic performance prediction model using correlation analysis (Spearman correlation, chi-square test, t-SNE), enhancing the analytical and predictive capabilities of each sub-model when processing input data.

[0079] like Figure 3 As described above, the present invention provides a method for operation Figure 2 The electronic device 30 for predicting academic performance shown may include a memory 301, a processor 302, and a bus, and may also include computer programs stored in the memory 301 and executable on the processor 302, such as the various functional modules of the academic performance prediction system.

[0080] The memory 301 includes at least one type of readable storage medium, such as a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. The memory 301 can be an internal storage unit of the electronic device 30 in some embodiments, such as a mobile hard disk of the electronic device 30. The memory 301 can also be an external storage device of the electronic device 30 in other embodiments, such as a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 30. Further, the memory 301 can include both an internal storage unit and an external storage device of the electronic device 30. The memory 301 can be used to store application software installed on the electronic device 30 and various types of data, such as the code of the academic performance prediction, and can also be used to temporarily store data that has been output or will be output.

[0081] The processor 302 can be composed of an integrated circuit in some embodiments, such as a single packaged integrated circuit or a plurality of packaged integrated circuits with the same or different functions, including one or more combinations of a central processing unit (CPU), a microprocessor, a digital processing chip, a graphics processor, and various control chips, etc. The processor 302 is a control unit of the electronic device 30, which connects various components of the entire electronic device 30 through various interfaces and lines, executes programs or modules stored in the memory 301 (such as a control program of the academic performance prediction system), and calls data stored in the memory 301, to execute various functions of the electronic device 30 and process data.

[0082] The processor 302 executes an operating system and various application programs installed on the electronic device 30. The processor 302 executes the application programs to implement the steps in the academic performance prediction method of the academic performance prediction system described above.

[0083] For example, the computer program can be divided into one or more modules, which are stored in the memory 301 and executed by the processor 302 to complete the present application. The one or more modules can be a series of computer program instructions capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device 30.

[0084] The integrated unit implemented in the form of the software function module can be stored in a computer readable storage medium, which can be nonvolatile or volatile. The software function module is stored in a storage medium and includes a plurality of instructions for enabling a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor to execute part of the functions of the method for predicting academic performance according to the embodiments of the present application.

[0085] In one embodiment, a storage medium is provided, and a computer program is stored in the storage medium. The computer program is executed by a processor to implement the steps implemented by the processor executing the computer program.

[0086] The above embodiments only illustrate the principles and effects of the present application, but are not used to limit the present application. Any person skilled in the art can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those skilled in the art without departing from the spirit and technical ideas disclosed by the present application should be covered by the claims of the present application.

Claims

1. A method of academic performance prediction, the method comprising: The method comprises: collecting first data related to academic performance, preprocessing the first data to obtain second data, wherein the preprocessing comprises data filtering and data encoding of the first data; converting the second data into input features, inputting the target model, and generating a prediction probability distribution corresponding to the academic performance; weighting and fusing the prediction probability distribution to generate a prediction result corresponding to the academic performance, and storing the prediction result in association with student identification information.

2. The method of claim 1, wherein, The first data includes at least one of family background data, psychological assessment data and academic performance data, wherein the data form of the psychological assessment data includes a symptom self-assessment scale, and the symptom self-assessment scale includes factors and item scores.

3. The method of claim 2, wherein, The preprocessing of the first data to obtain the second data comprises: eliminating psychological assessment data with a response time lower than a preset response time length.

4. The method of claim 3, wherein, Further comprising: determining the data type corresponding to the first data; for numerical data of the data type, encoding the numerical value of the first data; for categorical data of the data type, using integer category mapping for encoding, wherein the integer category mapping includes at least one of 0 / 1 encoding and continuous numerical encoding.

5. The method of claim 1, wherein, The conversion of the second data into input features, the inputting of the target model, and the generation of a prediction probability distribution corresponding to the academic performance comprise: mapping the numerical data included in the second data to a preset numerical range to obtain numerical features; performing embedding operation on the categorical data included in the second data to obtain categorical features, wherein the input features include the numerical features, the categorical features, and the combined features obtained by performing merging operation on the numerical features and the categorical features; inputting the numerical features, the categorical features and the combined features into at least one target model to generate the prediction probability distribution.

6. The method of claim 5, wherein, The target model includes at least one of sub-models for processing the numerical features, the categorical features and the combined features, and the weights of the sub-models are optimized and adjusted when the sub-models process the numerical features, the categorical features and the combined features.

7. The method of claim 6, wherein, Further comprising: evaluating the sub-models so that the weights of the sub-models corresponding to the same numerical features, categorical features and combined features are the same; performing correlation analysis on the sub-models to adjust the analysis-prediction ability of the sub-models on the numerical features, the categorical features and the combined features.

8. The method of claim 6, wherein, The weighting and fusing of the prediction probability distribution to generate a prediction result corresponding to the academic performance, and the storage of the prediction result in association with student identification information, comprise: receiving the prediction probability distribution from at least one sub-model; fusing and calculating the prediction probability distribution of each sub-model according to the fusion weight corresponding to the sub-model to obtain a weighted average probability as the prediction result; storing the prediction result and the student identification information in the form of key-value pairs.

9. An academic performance prediction system, comprising: The method comprises: The data acquisition module, the data preprocessing module, the training and testing module and the prediction module, wherein, The data acquisition module is used for collecting first data for academic performance, and the data preprocessing module is used for preprocessing the first data to obtain second data, wherein the preprocessing includes data filtering and data coding on the first data; The training and testing module is used for converting the second data into input features, inputting at least one target model, and generating a prediction probability distribution corresponding to the academic performance; The prediction module is used for weighted fusion of the prediction probability distribution to generate a prediction result corresponding to the academic performance, and the prediction result is associated with the student identification information and stored.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the academic performance prediction method according to any one of claims 1 to 8.

Images