Research on Academic Risk Early Warning Model Based on the Analysis of Learning Behavior Data

Accelerated by the rapid advancement of educational digital transformation, learners generate a wealth of process data during online learning. This study aims to employ educational data mining and machine learning methodologies to achieve early warnings for student academic risks. Utilizing the platform of Guangdong-Hong Kong-Macao Greater Bay Area University Online Open Course Alliance, the research collects and analyzes learning behavior data, extracting behavioral indicators closely related to student academic performance. Through an optimized BP neural network algorithm, a grade prediction model is constructed that uses student academic performance ranks as output features and determines input features via Spearman correlation analysis; hyper-parameters of the hidden layer are optimized using Python and MATLAB, followed by model evaluation and prediction, ultimately resulting in a student academic risk warning model. This study defines and divides student behavior data based on Flanders Interaction Analysis theory, extracting characteristic data with representational significance. Compared with other models, this model is more pertinent in summarizing learning behavior and predicting academic performance. Experimental results demonstrate that the model exhibits excellent performance in accuracy (93.63%), recall rate (93.71%), and F1-score (0.935). This model effectively identifies students at academic risk and provides targeted early warnings, thereby reducing their academic risks.