XAI-Based Student Performance Prediction: Peeling Back the Layers of LSTM and Random Forest’s Black Boxes

This paper focuses on enhancing the accuracy of student performance prediction using LSTM and Random Forest algorithms. These algorithms are trained on Jordan datasets and also aimed to shed light on their internal workings. The methods of LIME and SHAP for achieving explainability have been employed to gain insights into the inner mechanisms of these prediction models. The study found that different explanation techniques yielded diverse outcomes in identifying the crucial factors influencing student success, even when using the same group of students and machine learning models. The results revealed that the LSTM model was predominantly influenced by the Parent Answering Survey and behavior features, as revealed by the LIME and SHAP methodologies. Conversely, the SPAP and LIME techniques unveiled that the Students Absence Days and Behavior features had a more pronounced influence on explaining the outcomes of the RF model.