Academic risk early warning and intervention system and method thereof
By integrating multi-source data and using the ElasticNet regression model, a personalized academic risk early warning and intervention system was constructed. This system solves the problems of incomplete data and weak anti-interference ability of the model in academic risk management, and achieves efficient and automated academic risk management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-08
- Publication Date
- 2026-04-03
AI Technical Summary
Existing academic risk management systems rely on a single data source, resulting in incomplete assessment models, large biases in prediction results, weak model resistance to interference, rigid intervention mechanisms, lack of dynamic optimization, and inability to form an automated closed loop.
A multi-source data fusion terminal is used, combined with the ElasticNet regression model for feature selection and processing, to design personalized intervention strategies. A closed-loop system is formed through dynamic optimization units to achieve automated data collection, evaluation and intervention.
It has improved the accuracy and stability of academic risk assessment, achieved fully automated management, possesses self-iterative optimization capabilities, and improved management efficiency and response speed.
Smart Images
Figure CN121787912A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of educational informatization technology, and in particular to an academic risk early warning and intervention system and method. Background Technology
[0002] With the deepening of digitalization and the full credit system reform in higher education, academic risk management faces new challenges. Traditional academic early warning mechanisms mainly suffer from the following technical deficiencies: The data dimensions are limited, and the assessment model is one-sided: Existing systems rely heavily on explicit structured data such as academic performance, lacking integrated analysis of implicit unstructured data such as students' psychological traits and learning behaviors. Isolated data sources lead to incomplete input features for risk assessment models, resulting in biased prediction results.
[0003] Poor anti-interference ability and insufficient prediction stability: Traditional statistical models such as logistic regression are prone to overfitting when dealing with high-dimensional student feature data (such as multiple psychological traits and academic indicators) that may have multicollinearity. This results in the model performing well on the training set, but having weak generalization ability on new data and large fluctuations in warning accuracy.
[0004] The intervention mechanism is rigid and lacks a technological closed loop: existing assistance measures are mostly static and general recommendations that fail to be dynamically linked to risk assessment results and lack quantitative evaluation and feedback on intervention effectiveness. The coupling between different parts of the system (data collection, analysis, and intervention) is low, making it impossible to form an automated technological closed loop of "monitoring-assessment-intervention-reassessment," resulting in slow response and low efficiency.
[0005] Therefore, there is an urgent need for a technical solution that can integrate multi-source data, adopt robust prediction models, and achieve automated, precise intervention and dynamic optimization to improve the intelligence and effectiveness of academic risk management. Summary of the Invention
[0006] (a) Technical problems to be solved The present invention aims to address the deficiencies existing in the aforementioned background art, and the specific technical problems to be solved include: How can we use technology to automate the collection and effective integration of heterogeneous data from multiple sources, such as academic and psychological data, to provide a comprehensive and high-quality data foundation for risk assessment?
[0007] How can we construct an academic risk prediction model that can overcome feature collinearity interference and has strong generalization ability to improve the accuracy and stability of early warning?
[0008] How to design a closed-loop system architecture that can automatically generate and execute personalized intervention strategies based on risk assessment results, and continuously optimize models and strategies based on feedback data.
[0009] (II) Technical Solution To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides an academic risk early warning and intervention system based on multi-source data fusion and elastic network regression, characterized in that it includes: A data acquisition terminal group physically deployed in the campus network, the terminal group including: an academic data acquisition terminal directly connected to the academic affairs database server via a wired network interface, and a psychological assessment terminal equipped with an input device and a microcontroller, used to collect structured and unstructured raw student data; A central processing server, communicatively connected to the data acquisition terminal group, includes: The data fusion preprocessing unit is used to perform timestamp alignment, missing value imputation, and standardization and normalization on the received raw data to generate a structured feature dataset. The risk assessment engine is loaded with an offline-trained ElasticNet regression prediction model to process the structured feature dataset and output a quantified comprehensive risk value for students. The training process of the prediction model includes screening initial features based on the Pearson correlation coefficient and determining the optimal mixed parameters of L1 and L2 regularization terms through grid search and cross-validation. The strategy matching and generation unit has an embedded strategy rule library, which is used to compare the comprehensive risk value with multiple preset risk threshold ranges to determine the risk level, and automatically match and generate a set of personalized intervention strategies corresponding to the level. The warning and execution interface is used to automatically generate academic warning information when the risk level reaches high risk, and push the warning information and intervention strategy set to the associated academic affairs management platform or student mobile application through a predefined application programming interface protocol.
[0010] Secondly, the present invention provides a method for academic risk warning and intervention executed in a computing device, characterized in that it includes: Structured academic data from the academic affairs database and standardized psychological data from psychological assessment terminals are collected through a dedicated data interface protocol. Data cleaning and feature engineering are performed on the collected multivariate heterogeneous data, including time series alignment, categorical variable encoding, and continuous variable normalization, to form a unified feature vector. The feature vector is input into a pre-trained ElasticNet regression model to calculate a comprehensive score representing academic risk. The ElasticNet regression model is trained by applying L1 and L2 mixed regularization constraints to historical student data to solve feature multicollinearity and improve the model's generalization ability. The student's real-time risk level is determined by comparing the comprehensive score with a dynamically adjustable risk threshold. Based on the real-time risk level, personalized intervention strategies are retrieved and generated from a pre-built strategy knowledge graph, including at least one of psychological counseling, learning skills training, and curriculum adjustment. When the real-time risk level is high, an early warning mechanism is automatically triggered, generating a structured early warning report and sending it asynchronously to the relevant management interface via a message queue.
[0011] (III) Beneficial Effects Compared with the prior art, the technical solution provided by the present invention has the following technical effects: Improved accuracy and robustness of risk assessment: By adopting the ElasticNet regression model, which combines L1 and L2 regularization, it can effectively handle the multicollinearity problem in student feature data, automatically select features and constrain model complexity, thereby significantly reducing the risk of model overfitting and improving the stability of prediction results and generalization ability on new data.
[0012] The entire process is automated and intelligent: By designing a system architecture that includes dedicated data acquisition terminals, a central processing server and various functional units, and defining clear data interfaces and execution protocols, the entire chain of automated processing, from automatic acquisition of multi-source data, fusion preprocessing, intelligent risk assessment, to strategy matching and early warning triggering, is realized, which greatly reduces manual intervention and improves management efficiency and response speed.
[0013] An evolutionary, dynamically optimized closed loop was constructed: by setting up a model dynamic optimization unit and / or an effect evaluation unit, the system can incrementally learn the predictive model using new data generated after intervention, and optimize the intervention strategy library based on quantitative effect evaluation (such as a difference-in-differences model). This technological closed loop enables the system to have the ability to self-iterate and continuously improve, and can adapt to changes in the student population and academic environment over the long term. Attached Figure Description
[0014] Figure 1 This is a flowchart of the academic risk warning and intervention method provided in Embodiment 2 of the present invention. Detailed Implementation
[0015] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings and embodiments.
[0016] Example 1
[0017] This embodiment provides an academic risk early warning and intervention system based on multi-source data fusion and resilient network regression. The system is deployed within a university campus network and specifically includes: The data acquisition terminal group includes a high-performance industrial computer (APC) for academic data acquisition, located in the Academic Affairs Office computer lab. It connects directly to the core academic affairs database via a gigabit Ethernet interface and is configured with a scheduled task script to automatically extract structured data such as student course grades, GPA, and attendance from the previous day every morning. Simultaneously, a psychological assessment terminal (a customized tablet device with a built-in microcontroller) is deployed in the Psychological Counseling Center. After students complete standardized psychological scales (such as the Self-Regulation Efficacy Scale) on the device, the microcontroller packages and encrypts the answer data along with a precise timestamp.
[0018] Central Processing Server: An enterprise-grade server is used, connected to the aforementioned terminals via the campus network. The system software running on the server includes: Data fusion preprocessing unit: Receives data streams from both terminals. First, it timestamps the academic data records and psychological assessment data packets using student ID and time as keys. Then, it performs Z-score standardization on continuous variables (such as grades and scores) and one-hot encoding on categorical variables (such as majors). For a small number of missing values, it imputes them using the mean of students in the same class.
[0019] Risk assessment engine: The core loads a pre-trained ElasticNet regression model. The training process for this model has been completed historically: Data from the past three cohorts of students was collected. First, Pearson correlation coefficients were calculated between each feature and academic failure (e.g., failing a course), and features with a significance level (p-value) less than 0.05 were selected. Using the selected feature set, the ElasticNetCV function was called from Python's scikit-learn library. Through 10-fold cross-validation and grid search, the optimal regularization parameters α (controlling the overall regularization strength) and l1_ratio (controlling the proportion of L1 regularization) that minimize the model's mean squared error were determined. The trained model parameters (coefficients, intercepts) are persistently stored in a server-side in-memory database.
[0020] Strategy Matching and Generation Unit: This unit embeds a strategy rule base, stored in key-value pairs. For example, a rule might be defined as: "risk_level": "high", "threshold": 0.7, "actions": ["Trigger crisis intervention process", "Notify counselor", "Recommend leave of absence buffer application"]. This unit matches the corresponding rule based on the risk value (between 0 and 1) output by the engine.
[0021] Early warning and execution interface: Designed using RESTful API, it interfaces with the school's existing academic affairs management system and student affairs system. When an early warning is needed, the system automatically constructs a JSON data packet conforming to the interface specification (containing student ID, risk value, warning level, and a list of recommended measures), and sends it to the target system via an HTTP POST request, triggering its internal workflow.
[0022] The system's workflow is as follows: Data collection and uploading: The terminal collects data and uploads it to the server according to the set period.
[0023] Fusion and Evaluation: The preprocessing unit completes data cleaning and feature engineering, generating the current feature vector. The risk assessment engine calls the ElasticNet model to calculate the comprehensive risk value R (e.g., 0.72).
[0024] Strategy matching: The strategy matching unit determines that if R ≥ 0.7, it matches the "high-risk" rule.
[0025] Early Warning and Execution: The early warning interface packages high-risk warnings and associated "crisis intervention procedures" into a single package and pushes it to the academic affairs system API. Upon receiving the notification, the academic affairs system automatically generates a high-priority to-do item on the counselor's workbench.
[0026] Dynamic optimization (background process): Each week, the model dynamic optimization unit collects follow-up data (if any) from all students who received intervention in the past week, and uses this data as a micro-batch to perform an incremental learning round on the ElasticNet model, fine-tuning the model parameters. At the end of each semester, the effectiveness evaluation unit runs a difference-in-differences model to analyze the net effect of each intervention strategy in the current period, and adjusts some recommended measures or risk thresholds in the strategy rule base accordingly.
[0027] Example 2
[0028] refer to Figure 1 This embodiment provides a method for academic risk warning and intervention executed on a computing device. This method can be executed by the system in Embodiment 1 or deployed on a cloud service platform. The method steps include: S101: Multi-source data acquisition. Regularly retrieve academic data by configuring a database connector (such as JDBC); obtain encrypted assessment result data packets by calling the API provided by the psychological assessment terminal.
[0029] S102: Data Cleaning and Feature Engineering. Perform the following on the retrieved data: remove obviously invalid records (such as assessments with excessively short answer times); align the "exam date" of academic data with the "submission time" of psychological assessments to ensure consistent data time windows; normalize numerical features and encode categorical features such as "major" and "grade".
[0030] S103: Risk Assessment Based on the ElasticNet Model. The processed feature vectors are input into the trained ElasticNet model. Due to its regularization properties, this model is insensitive to collinearity in the input features and can output stable risk probability values. For example, the model formula can be simplified to: Risk Value = sigmoid(β0 + β1 * feature1 + s… + βn * featuren + λ1 * L1_penalty + λ2 * L2_penalty).
[0031] S104: Risk Classification and Policy Matching. The system maintains a dynamic threshold table (e.g., {'Low':[0,0.3),'Medium':[0.3,0.7),'High':[0.7,1.0]}). The risk values obtained in S203 are matched to the corresponding levels, and the policy generation function corresponding to that level is activated.
[0032] S105: Personalized Intervention Plan Generation. The strategy generation function combines specific suggestions from the knowledge graph based on the risk level and details in the feature vector (such as the psychological trait weakness being "self-control"). For example, for a student who is "high-risk" and has "low self-control," the generated plan includes "suggesting a weekly mindfulness counseling session" and "recommending the use of a time management app."
[0033] S106: Automatic Early Warning and Task Distribution. When the level is "High," the method automatically calls a messaging service (such as RabbitMQ) to publish the early warning information and detailed solutions to the designated message queue. The student affairs system and the psychological counseling system subscribe to this queue, each obtaining relevant information and creating internal tasks.
[0034] S107: Effect Feedback and Model Iteration. The method periodically (e.g., every two months) executes the optimization sub-process: collecting subsequent academic performance data of the intervened students, adding them as new samples to the training set, and initiating a round of incremental model training (Retraining) to maintain the model's timeliness.
[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. An academic risk early warning and intervention system, characterized in that, include: A data acquisition terminal group physically deployed in the campus network, the terminal group including: an academic data acquisition terminal directly connected to the academic affairs database server via a wired network interface, and a psychological assessment terminal equipped with an input device and a microcontroller, used to collect structured and unstructured raw student data; A central processing server, communicatively connected to the data acquisition terminal group, includes: The data fusion preprocessing unit is used to perform timestamp alignment, missing value imputation, and standardization and normalization on the received raw data to generate a structured feature dataset. The risk assessment engine is loaded with an offline-trained ElasticNet regression prediction model to process the structured feature dataset and output a quantified comprehensive risk value for students. The training process of the prediction model includes screening initial features based on the Pearson correlation coefficient and determining the optimal mixed parameters of L1 and L2 regularization terms through grid search and cross-validation. The strategy matching and generation unit has an embedded strategy rule library, which is used to compare the comprehensive risk value with multiple preset risk threshold ranges to determine the risk level, and automatically match and generate a set of personalized intervention strategies corresponding to the level. The warning and execution interface is used to automatically generate academic warning information when the risk level reaches high risk, and push the warning information and intervention strategy set to the associated academic affairs management platform or student mobile application through a predefined application programming interface protocol.
2. The system according to claim 1, characterized in that, The data fusion preprocessing unit performs standardization and normalization processing, specifically using the Z-score standardization method.
3. The system according to claim 1, characterized in that, The central processing server also includes a model dynamic optimization unit, which is used to periodically collect new student data after intervention as an incremental dataset to perform incremental training and parameter updates on the ElasticNet regression prediction model.
4. The system according to claim 1, characterized in that, The intervention strategies pre-stored in the strategy rule base include at least one of the following for different risk levels: automatically pushing learning resource links to students with mild risk. Generate and assign one-on-one psychological counseling appointments and course adjustment suggestions for students at moderate risk; For students at high risk, a crisis intervention process involving multiple departments such as academic affairs, student affairs, and psychological counseling is triggered.
5. The system according to claim 1, characterized in that, The system also includes an effect evaluation unit, which collects students' subsequent academic and psychological data after the intervention period ends, and uses a difference-in-differences model to calculate the average treatment effect of the intervention strategy set in order to optimize the strategy rule base.
6. The system according to claim 1, characterized in that, The data collected by the psychological assessment terminal includes scores on self-regulation efficacy, environmental adaptability, and intrinsic motivation factors obtained through standardized psychological scales.
7. The system according to claim 1, characterized in that, The academic data collection terminal periodically collects data including course grades, grade points, attendance records, and national English proficiency test scores.
8. A method for academic risk early warning and intervention as described in claim 1, characterized in that, include: Structured academic data from the academic affairs database and standardized psychological data from psychological assessment terminals are collected through a dedicated data interface protocol. Data cleaning and feature engineering are performed on the collected multivariate heterogeneous data, including time series alignment, categorical variable encoding, and continuous variable normalization, to form a unified feature vector. The feature vector is input into a pre-trained ElasticNet regression model to calculate a comprehensive score representing academic risk. The ElasticNet regression model is trained by applying L1 and L2 mixed regularization constraints to historical student data to solve feature multicollinearity and improve the model's generalization ability. The student's real-time risk level is determined by comparing the comprehensive score with a dynamically adjustable risk threshold. Based on the real-time risk level, personalized intervention strategies are retrieved and generated from a pre-built strategy knowledge graph, including at least one of psychological counseling, learning skills training, and curriculum adjustment. When the real-time risk level is high, an early warning mechanism is automatically triggered, generating a structured early warning report and sending it asynchronously to the relevant management interface via a message queue.
9. The method according to claim 8, characterized in that, The dynamically adjustable risk threshold is adjusted periodically through feedback loops based on historical intervention effect data, with the goal of optimizing the accuracy of early warnings.
10. The method according to claim 8, characterized in that, After generating a personalized intervention strategy, the method further includes: writing the executable tasks in the intervention plan into the to-do list of the school's office automation system through the dedicated data interface protocol, and tracking the task completion status.