Deep learning-based student score prediction method and system

Abstract

The invention discloses a student score prediction method and system based on deep learning, and relates to the technical field of education, and the system comprises a data collection module, a data preprocessing module, a model training and optimization module, a prediction and analysis module, a user interface (UI) module, and a feedback and iteration improvement module. And a plurality of deep learning models are used to learn a complex relationship between features and scores, and accurate prediction of future academic scores of students is realized. According to the student score prediction method and system based on deep learning, a set of student score prediction system which is high in precision, explainable and dynamically evolved is constructed by fusing multi-source education data and a deep learning technology, so that the problems of low prediction precision and insufficient data utilization of a traditional method are solved; in addition, timeliness and accuracy of academic prediction are effectively improved, personalized education and educational resource optimization configuration are assisted, and data driving and scientization of educational decision are promoted.

Description

Technical Field

[0001] This invention relates to the field of educational technology, specifically to a method and system for predicting student performance based on deep learning. Background Technology

[0002] With the development of big data and artificial intelligence technologies, the education field has begun to explore how to improve teaching quality and student learning efficiency through technological means. Among these, student performance prediction is an important research direction, which aims to help class teachers / counselors and teachers in universities identify students who may have learning difficulties in advance, so as to provide necessary support and intervention in a timely manner.

[0003] However, in existing technologies, predictive models often rely on traditional statistical methods, such as linear regression, logistic regression, and multivariate statistical analysis. These methods depend on strict assumptions (such as linear data distribution and feature independence), making it difficult to handle the nonlinear and high-dimensional characteristics prevalent in educational data. Consequently, their predictive accuracy is limited, and they struggle to accurately assess the complex learning patterns and behavioral characteristics. Furthermore, existing systems generally lack closed-loop optimization mechanisms, making it difficult to dynamically adjust models based on new data and intervention effects after training, leading to a long-term decline in predictive accuracy. At the same time, the interpretability of prediction results is insufficient, as homeroom teachers / counselors and other teachers cannot clearly understand the specific academic risks faced by a particular student, affecting the targeted nature of intervention measures.

[0004] To address the aforementioned issues, there is an urgent need for innovative design based on existing technologies. Therefore, we propose a method and system that uses deep learning algorithms to analyze various academic data of students and predict their future academic performance, which can effectively solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a student performance prediction method and system based on deep learning, so as to solve the problems of insufficient accuracy, inadequate data utilization, and lack of dynamic optimization of the model in the existing student performance prediction methods mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a student performance prediction system based on deep learning, comprising a data collection module, a data preprocessing module, a model training and optimization module, a prediction and analysis module, a user interface (UI) module, and a feedback and iterative improvement module;

[0007] Data collection module: Connects to the student learning management system (LMS) of various universities via interfaces to collect student academic data and student access data;

[0008] Data preprocessing module: Cleans and denoises the collected data, and performs necessary feature engineering, such as normalization and encoding categorical variables.

[0009] Model training and optimization module: Build and train deep learning models, use historical data for supervised learning, and adjust network parameters through backpropagation algorithm;

[0010] Prediction and Analysis Module: Uses a trained model to predict students' future performance and provides possible explanations and analyses;

[0011] User Interface (UI) module: Provides educators with an interactive interface to display prediction results and related analysis information. Educators and department heads can track and provide feedback to students with suspected academic difficulties and help them solve learning problems in a targeted manner.

[0012] Feedback and Iterative Improvement Module: Based on the difference between actual results and predicted results, new data is collected and the model is updated. In addition, based on the results of educators' follow-up visits to students, the accuracy of predictions is continuously improved.

[0013] Furthermore, the data collection module adopts a multi-source interface approach to achieve automated collection of full-volume educational data, wherein each data item specifically includes:

[0014] Basic data: By connecting to the school's unified identity authentication system, statistical data such as student gender, grade, major, entrance examination scores, and family economic status are collected;

[0015] Academic data: By connecting with the university's academic affairs office-student learning management system (LMS), we collect structured data such as students' exam scores, credits earned, grade points, course schedule information, homework submission records, and online learning behavior logs for each semester;

[0016] Campus access data: By connecting to the capture system of campus gates, panel machines and other equipment, behavioral data such as student access records are collected.

[0017] Furthermore, the data preprocessing module includes processing abnormal data by combining automated rules with manual verification during data cleaning. For missing values, continuous features are filled using the KNN algorithm, and categorical features are filled using the mode. For outliers, extreme values ​​are identified and removed using the Z-score method, and the latest data is retained for duplicate records to ensure the accuracy of the final data.

[0018] Furthermore, when constructing and training the deep learning model, the model training and optimization module, based on the multimodal characteristics of educational data, constructs a hybrid model architecture that integrates multiple neural networks, specifically including:

[0019] Basic model selection: Multilayer perceptron (MLP) was selected to process structured comprehensive features; convolutional neural network (CNN) was selected to extract local correlation patterns in performance data; recurrent neural network (RNN) was selected and LSTM (Long Short-Term Memory) was used to process time series data.

[0020] Model fusion strategy: A feature-level fusion approach is adopted, which concatenates the structured features of MLP, the local pattern features of CNN, and the temporal features of LSTM into a unified feature vector, which is then connected to the final decision layer (a 2-layer fully connected network). The final prediction result is output through a weighted voting mechanism to improve the robustness of the model.

[0021] Furthermore, the training unit in the model training and optimization module specifically includes:

[0022] Dataset partitioning: Time series partitioning is used to ensure consistent data distribution;

[0023] Training process: Model training is implemented based on the TensorFlow framework, with Adam selected as the optimizer and cross-entropy loss as the loss function;

[0024] Validation method: Five-fold cross-validation was used to evaluate the model stability;

[0025] Evaluation metrics: Accuracy, precision, recall, F1-score, and root mean square error (RMSE) are used to evaluate model performance to ensure the comprehensiveness and reliability of prediction results.

[0026] Furthermore, the prediction and analysis module specifically includes:

[0027] Results presentation: Prediction results are presented in a multi-dimensional dashboard format through the user interface (UI) module, and the user interface (UI) supports access from multiple terminals such as PC, APP and H5, increasing the ease of operation;

[0028] Interpretability analysis: The SHAP value algorithm is used to calculate feature importance, generate a feature influence heatmap, clarify the contribution of each key factor to the prediction results, and help educators understand the prediction logic;

[0029] Risk Classification: Based on the combination of predicted grades and risk characteristics, students are classified into three risk levels: high-level abnormality, which is reported to the Dean of Student Affairs and the relevant leaders; medium-level abnormality, which is reported to the Department Head and the Director of Student Affairs; and low-level abnormality, which is reported to the class teacher / counselor or the course instructor.

[0030] Furthermore, when the curriculum system is adjusted or the teaching model is updated, the feedback and iterative improvement module adopts a combination of incremental training and full retraining to establish a closed-loop feedback mechanism. Incremental training is carried out monthly (only new data is used to update model parameters), and full retraining is carried out every semester (based on all historical data) to ensure that the model adapts to changes in learning patterns.

[0031] Furthermore, the feedback and iterative improvement module also incorporates an optimization mechanism that calculates prediction bias based on feedback data, adjusts feature weights through reinforcement learning, and periodically evaluates model performance. When the accuracy drops by more than 5%, it triggers model structure optimization, such as increasing the number of neurons in the hidden layer.

[0032] A deep learning-based method for predicting student grades includes the following steps:

[0033] Step 1: Data Collection and Preprocessing

[0034] It integrates with the LMS system and the campus access control system, and preprocesses the acquired data to remove invalid or abnormal data, fill in missing values, and convert non-numerical data into a format acceptable to machine learning algorithms.

[0035] Step 2: Feature engineering, model selection and training, and model evaluation and optimization.

[0036] Based on educational psychology and learning theory, key features affecting student performance are selected and feature vectors are constructed. Appropriate deep learning models, such as convolutional neural networks (CNN), recurrent neural networks (RNN), or ensembles of multiple models, are selected. The performance of the models is evaluated through cross-validation, and the model structure and hyperparameters are adjusted based on the evaluation results.

[0037] Step 3: Predict the output through the user interface (UI)

[0038] The final prediction results are presented to educators through the user interface (UI) module, along with detailed reports and explanations to help them understand the reasons behind the predictions. Educators can then follow up and provide feedback to students with suspected academic difficulties, and offer targeted assistance to help them overcome their learning challenges.

[0039] Step 4: Feedback Collection and Model Iteration

[0040] We collect feedback and suggestions from educators and new academic data to continuously iterate and update the model in order to improve the long-term stability and predictive accuracy of the system.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This deep learning-based student performance prediction method and system integrates multi-source educational data with deep learning technology to build a high-precision, interpretable, and dynamically evolving student performance prediction system. It not only solves the problems of low prediction accuracy and insufficient data utilization in traditional methods, but also effectively improves the timeliness and accuracy of academic prediction, helps personalized education and the optimal allocation of educational resources, and promotes data-driven and scientific educational decision-making.

[0043] 1. Significantly improved prediction accuracy: Through multi-source data fusion and deep learning models, the system can capture complex learning patterns that are difficult to identify using traditional methods, achieving a prediction accuracy of 89%, which effectively improves the accuracy and reliability of student performance prediction.

[0044] 2. Easy to operate and highly scalable: The layered architecture design makes the system easy to deploy and expand. It supports seamless integration with existing LMS, campus card and other systems without the need to modify the original education IT architecture. The user interface is simple and intuitive, and educators can use it without professional data knowledge, which lowers the technical threshold and makes it easy to operate and expand.

[0045] 3. Achieve early and precise intervention: Based on 24 / 7 dynamic data collection and real-time analysis, the system can trigger an early warning when students experience learning difficulties (such as skipping classes for two consecutive weeks or a sudden drop in study time). Compared with the traditional "post-study performance analysis", it can detect risks 4-8 weeks in advance, making it easier for educators to intervene in a timely manner.

[0046] 4. Support for personalized educational decisions: Through feature importance analysis and interpretability reports, educators can identify the root causes of students' academic problems and develop targeted measures. At the same time, the system can also provide schools with resource optimization suggestions, which helps educators personalize teaching and rationally allocate school educational resources, thereby improving the efficiency of educational resource allocation.

[0047] 5. Promote the scientific development of digital education: Establishing a closed-loop feedback mechanism can ensure that the model can adapt to changes in the educational environment. When the curriculum system is adjusted or the teaching model is updated, incremental training can quickly adapt to the new data distribution, solve the problem of "static aging" of traditional models, and promote data-driven and scientific educational decision-making. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the system architecture of the present invention.

[0049] Figure 2 This is a schematic diagram of the risk assessment and early warning logic of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1: Please refer to Figure 1 The present invention provides the following technical solution: a student performance prediction system based on deep learning, comprising a data collection module, a data preprocessing module, a model training and optimization module, a prediction and analysis module, a user interface (UI) module, and a feedback and iterative improvement module;

[0052] Data collection module: It connects to the student learning management system (LMS) of various universities through interfaces to collect data on each student, exam scores for each semester, online learning behavior logs, homework submission records, and student access data captured by campus gates, panel machines, etc.

[0053] Data preprocessing module: Cleans and denoises the collected data, and performs necessary feature engineering, such as normalization and encoding categorical variables.

[0054] Model training and optimization module: Build and train deep learning models, use historical data for supervised learning, and adjust network parameters through backpropagation algorithm;

[0055] Prediction and Analysis Module: Uses a trained model to predict students' future performance and provides possible explanations and analyses;

[0056] User Interface (UI) module: Provides educators with an interactive interface to display prediction results and related analysis information. Educators and department heads can track and provide feedback to students with suspected academic difficulties and help them solve learning problems in a targeted manner.

[0057] Feedback and Iterative Improvement Module: Based on the difference between actual results and predicted results, new data is collected and the model is updated. In addition, based on the results of educators' follow-up visits to students, the accuracy of predictions is continuously improved.

[0058] In summary, the system automatically extracts key features from existing educational data (exam data for each semester, including earned credits and GPA), student behavior data, and campus access snapshot data 24 / 7. It then uses a deep learning model to learn the complex relationship between these features and student performance, thereby achieving accurate predictions of future academic results.

[0059] The prediction and analysis module specifically includes:

[0060] Results presentation: The prediction results are presented in the form of multi-dimensional dashboards through the user interface (UI) module, including academic dashboard (individual performance trend, class ranking prediction), truancy dashboard (high frequency truancy statistics), and failing dashboard (high risk failing course warning). The user interface (UI) supports access from multiple terminals such as PC, APP and H5, increasing the ease of operation.

[0061] Interpretability analysis: The SHAP value algorithm is used to calculate feature importance and generate a feature influence heatmap to clarify the contribution of key factors such as "insufficient learning time", "high truancy rate" and "high course difficulty" to the prediction results, helping educators understand the prediction logic;

[0062] Risk Classification: Based on the combination of predicted grades and risk characteristics, students are divided into three risk levels: High-level anomaly: predicted failure of ≥3 courses and on-campus study time <40% according to data, reported to the Dean of Student Affairs and relevant leaders; Medium-level anomaly: predicted failure of 1-2 courses or absence rate ≥30%, reported to the Department Head and Dean of Student Affairs; Low-level anomaly: predicted grade of <60 in a single course or ≥5 delayed assignment submissions, reported to the class teacher / counselor or course instructor, to achieve targeted graded early warning and promote timely intervention by educators.

[0063] Example 2:

[0064] Based on Example 1, a multi-source interface approach for the data collection module is also disclosed to achieve automated collection of full-volume educational data. The specific data items include:

[0065] Basic data: By connecting to the school's unified identity authentication system, statistical data such as student gender, grade, major, entrance examination scores, and family economic status are collected;

[0066] Academic data: By connecting with the university's academic affairs office-student learning management system (LMS), we collect structured data such as students' exam scores (including midterm, final, and in-class quizzes), credits earned, grade points, course schedule information, homework submission records (including submission time and quality score), and online learning behavior logs (login frequency, video viewing time, and resource downloads) for each semester.

[0067] Campus access data: By connecting to the capture system of campus gates, panel cameras and other equipment, we collect behavioral data such as student access records on campus (time of entering and leaving dormitories, time of entering and leaving teaching buildings, length of stay in the library, and distribution of campus activity areas).

[0068] Furthermore, the data preprocessing module includes handling abnormal data by combining automated rules with manual verification during data cleaning. For missing values, continuous features (such as learning duration) are filled using the KNN algorithm, and categorical features (such as major) are filled using the mode. For outliers, extreme values ​​are identified and removed using the Z-score method (|Z|>3), and the latest data is retained for duplicate records to ensure the accuracy of the final data.

[0069] Furthermore, when building and training deep learning models, the model training and optimization module constructs a hybrid model architecture that integrates multiple neural networks, based on the multimodal characteristics of educational data. Specifically, this includes:

[0070] The basic model selection is as follows: A Multilayer Perceptron (MLP) is chosen to process structured comprehensive features, employing a three-layer hidden layer design: 128 neurons in the first layer, 64 neurons in the second layer, and 32 neurons in the third layer. The ReLU activation function is used to address the vanishing gradient problem. The output layer uses the Softmax activation function to output the predicted grade level (Excellent / Good / Pass / Fail). A Convolutional Neural Network (CNN) is chosen to extract local correlation patterns from the grade data. The input layer is a course grade matrix (rows represent courses, columns represent grade dimensions). Two convolutional layers extract the correlation features between courses, which are then reduced in dimensionality by a MaxPooling layer before being connected to a fully connected layer. A Recurrent Neural Network (RNN) is chosen, using an LSTM (Long Short-Term Memory) network to process time-series data. The input is sequential data such as weekly study time and access records. Three LSTM units capture long-term dependencies (such as the trend of learning status changes within a semester), outputting a time-series feature vector.

[0071] Model fusion strategy: A feature-level fusion approach is adopted, which concatenates the structured features of MLP, the local pattern features of CNN, and the temporal features of LSTM into a unified feature vector, which is then connected to the final decision layer (a 2-layer fully connected network). The final prediction result is output through a weighted voting mechanism to improve the robustness of the model.

[0072] Meanwhile, the training units of the model training and optimization module specifically include:

[0073] Dataset partitioning: A time series partitioning method is adopted, such as using the data from the first 3 academic years as the training set (70%), the data from the first half of the 4th academic year as the validation set (15%), and the data from the second half of the 4th academic year as the test set (15%), to ensure the consistency of data distribution;

[0074] Training process: Model training is implemented based on the TensorFlow framework. The optimizer is Adam (initial learning rate 0.001, decaying by 10% every 5 rounds). The loss function is cross-entropy loss, the batch size is set to 32, and the number of iterations is 50 rounds.

[0075] Validation method: Five-fold cross-validation was used to evaluate the stability of the model. Each validation was performed by randomly dividing the data into five parts, with four parts used for training and one part for validation. This was repeated five times and the average performance index was calculated.

[0076] Evaluation metrics: Accuracy, precision, recall, F1-score, and root mean square error (RMSE) are used to evaluate model performance to ensure the comprehensiveness and reliability of prediction results.

[0077] When the curriculum is adjusted or the teaching model is updated, the feedback and iterative improvement module adopts a combination of incremental training and full retraining to establish a closed-loop feedback mechanism. Incremental training is conducted monthly (updating model parameters only with new data), and full retraining is conducted every semester (retraining based on all historical data) to ensure that the model adapts to changes in learning patterns. Furthermore, the feedback and iterative improvement module also has an optimization mechanism that calculates prediction bias based on feedback data and adjusts feature weights through reinforcement learning. For example, for students whose grades have significantly improved after intervention, the feature weight of "tutoring participation" is increased. The model performance is evaluated regularly, and when the accuracy drops by more than 5%, model structure optimization is triggered, such as increasing the number of neurons in the hidden layer.

[0078] Example 3:

[0079] Based on Example 2, a student performance prediction method based on deep learning is also disclosed, including the following steps:

[0080] Step 1: Data Collection and Preprocessing

[0081] It integrates with the LMS system (to obtain students' historical academic data, including but not limited to login frequency, study duration, discussion forum interaction, and assignment submission quality) and the campus access system (to obtain students' on-campus access snapshot data). It also preprocesses the acquired data, removes invalid or abnormal data, fills in missing values, and converts non-numerical data into a format acceptable to machine learning algorithms.

[0082] Step 2: Feature engineering, model selection and training, and model evaluation and optimization.

[0083] Based on educational psychology and learning theory, key features that influence student performance are selected, such as learning habits, time management ability, and participation. Feature vectors are constructed, and appropriate deep learning models are selected, such as convolutional neural networks (CNNs) to discover local patterns in performance data, recurrent neural networks (RNNs) to process time-series data, or multiple models are integrated to improve robustness and accuracy. The performance of the models is evaluated through cross-validation, and the model structure and hyperparameters are adjusted based on the evaluation results.

[0084] Step 3: Predict the output through the user interface (UI)

[0085] The final prediction results are presented to educators through the user interface (UI) module, along with detailed reports and explanations, to help educators understand the reasons behind the prediction results. Educators and department heads can then follow up and provide feedback to students with suspected academic difficulties, offering targeted assistance to help them overcome learning challenges.

[0086] Step 4: Feedback Collection and Model Iteration

[0087] We collect feedback and suggestions from educators and new academic data to continuously iterate and update the model in order to improve the long-term stability and predictive accuracy of the system.

[0088] Example 4:

[0089] In one specific embodiment, we used student data from the School of Mechanical Engineering at a university as our dataset. The data included students' online learning activity records, assignment grades, midterm and final exam scores, and campus access data throughout the semester. We employed a Multilayer Perceptron (MLP) with two hidden layers as the prediction model and used the TensorFlow framework for model training and testing. After multiple rounds of parameter tuning and model optimization, our system achieved an accuracy of 89% on the test set, significantly higher than the prediction results of traditional statistical methods.

[0090] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0091] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based student performance prediction system, characterized by: The system includes a data collection module, a data preprocessing module, a model training and optimization module, a prediction and analysis module, a user interface (UI) module, and a feedback and iterative improvement module. The data collection module collects student academic data and student access data from the university student management system (LMS) through an interface. The data preprocessing module cleans and denoises the collected data, and performs necessary feature engineering such as normalization and encoding categorical variables. The model training and optimization module builds and trains a deep learning model using historical data for supervised learning and adjusts network parameters through backpropagation algorithm. The prediction and analysis module uses the trained model to predict future student performance and provides possible explanations and analysis. The user interface (UI) module provides an interactive interface for educators to display prediction results and related analysis information. The feedback and iterative improvement module collects new data and updates the model based on the differences between actual performance and predicted results, and continuously improves prediction accuracy based on educator tracking and visiting results. 2.The deep learning-based student performance prediction system of claim 1, wherein: The data collection module uses a multi-source interface docking method to automatically collect full-quantity education data, which includes: Basic data: Collects student gender, grade, major, admission score, family economic status, and other statistical data by docking with the school unified identity authentication system. Academic data: Collects student exam scores, earned credits, GPA, course schedule information, assignment submission records, and online learning behavior logs by docking with the university student management system (LMS). Campus access data: Collects student access records and other behavior data by docking with campus gate and panel machine snapshot systems. 3.The deep learning-based student performance prediction system of claim 2, wherein: The data preprocessing module uses automated rules and manual verification to handle abnormal data during data cleaning. For missing values, KNN algorithm is used for continuous features and mode filling for categorical features. For outliers, Z-score method is used to identify and remove extreme values, and the latest data is retained for duplicate records. 4.The deep learning-based student performance prediction system of claim 3, wherein: The model training and optimization module builds a hybrid model architecture that integrates multiple neural networks based on the multi-modal characteristics of education data. Specifically, it includes: Base model selection: Selects multi-layer perceptron (MLP) to process structured comprehensive features, convolutional neural network (CNN) to extract local correlation patterns in score data, and recurrent neural network (RNN) with LSTM (long short-term memory network) to process time series data. Model fusion strategy: Uses feature-level fusion to concatenate MLP structured features, CNN local pattern features, and LSTM time series features into a unified feature vector, which is input into the final decision layer to output the final prediction result through weighted voting mechanism. 5.The deep learning-based student performance prediction system of claim 4, wherein: The training unit of the model training and optimization module includes: Data set division: Uses time series division method to ensure data distribution consistency. Training process: Model training is implemented based on the TensorFlow framework, the optimizer is Adam, and the loss function is cross-entropy loss; Verification method: 5-fold cross-validation is used to evaluate the stability of the model; Evaluation index: The accuracy, precision, recall, F1-score, and root mean square error (RMSE) are used to evaluate the performance of the model. 6.The deep learning-based student performance prediction system of claim 5, wherein: The prediction and analysis module specifically includes: Result display: The prediction results are presented in the form of multi-dimensional dashboards through the user interface (UI) module, and the user interface (UI) supports PC, APP, and H5 multi-terminal access; Interpretability analysis: SHAP value algorithm is used to calculate feature importance and generate feature influence heat maps to clearly show the contribution of key factors to the prediction results, helping educators understand the prediction logic; Risk classification: According to the combination of predicted results and risk characteristics, students are divided into three levels of risk: high abnormality, and pushed to the school labor department director and division leaders; medium abnormality, and pushed to the department director and labor director; low abnormality, and pushed to the class teacher / counselor or teacher. 7.The deep learning-based student performance prediction system of claim 6, wherein: When the curriculum system is adjusted and the teaching mode is updated, the feedback and iterative improvement module uses a combination of incremental training and full retraining to establish a closed-loop feedback mechanism, with incremental training every month and full retraining every semester. 8.The deep learning based student performance prediction system of claim 7, wherein: The feedback and iterative improvement module also has an optimization mechanism that calculates prediction bias based on feedback data and adjusts feature weights through reinforcement learning; the model performance is regularly evaluated, and model structure optimization is triggered when the accuracy drops by more than 5%. 9.A deep learning-based student performance prediction method applied to the deep learning-based student performance prediction system of claims 1-8, characterized in that: The steps include: Step one: data collection and preprocessing Integrate with LMS system, campus access system, and preprocess the obtained data to remove invalid or abnormal data, fill in missing values, and convert non-numeric data into a format acceptable to machine learning algorithms; Step two: build feature engineering, select and train the model, and evaluate and optimize the model Based on educational psychology and learning theory, select key features that affect student performance, build feature vectors, and select appropriate deep learning models such as convolutional neural networks (CNN), recurrent neural networks (RNN), or integrate multiple models, evaluate the performance of the model through cross-validation, and adjust the model structure and hyperparameters based on the evaluation results; Step three: prediction output through user interface (UI) Present the final prediction results to educators through the user interface (UI) module and provide detailed reports and explanations to help understand the reasons behind the prediction results, and educators track feedback on suspected academic abnormal students to help students solve learning difficulties; Step four: feedback collection and model iteration Collect educator feedback and new academic data to continuously update the model to improve the long-term stability and prediction accuracy of the system.

Images