LSTM‐Based Student Performance Prediction Model Incorporating Peer Effects and Explained Through SHAP

Everyone pays attention to the quality of education, and the performance of students in the learning process is an important manifestation of the quality of education. Many previous studies have applied AI technologies, especially deep learning models, to mine students′ learning process data to predict students′ final performance and provide early warnings for students with poor academic performance. However, no researchers have paid attention to the peer effects. This oversight leads to inaccurate models and makes early warnings difficult to implement. In addition, there have been relatively few previous studies that have conducted interpretability analysis on the model, resulting in less transparent and credible results. These gaps inspired the development of this research. This paper proposes a student performance prediction model based on long short‐term memory (LSTM) guided by the peer effects principle, named long short‐term memory‐peer effects (LSTM‐PE), which introduces a peer layer to improve the basic LSTM model. The model finally applies an open‐source library, named SHapley Additive exPlanations (SHAP), to interpret the results in order to better identify students of different performance categories. LSTM‐PE can achieve an accuracy of more than 90% in the first few weeks, providing more effective support for early warnings. The results of the interpretation showed that regular participation in learning activities and review after class can promote academic performance and even lead students to excellence.