Leveraging Explainable AI to Analyze and Improve Student Engagement in Virtual Learning Environment

Student behavior in online learning reflects their level of engagement with the learning process. Student engagement predictions are conducted to identify students with low engagement so that interventions can be implemented to improve it, as lack of engagement can affect final grades, material retention, and dropout rates. However, the system is still limited to providing prediction results because the machine learning architecture used is a "Black Box". Although accurate predictions, high accuracy is not the only important factor in the model's success. Trust in the model is also crucial, where students and teachers need to understand the reasoning behind the predictions so that the results can be explained and interpreted. This research uses three classification models, namely Random Forest, XGBoost, and SVM, with prediction results ranging from 97% to 99%. The Explainable AI (XAI) techniques used are SHAP and LIME. SHAP focuses on global explanations, while LIME focuses on local explanations. The results show that the explanations using SHAP and LIME are consistent both globally and locally for Random Forest and XGBoost, but slightly different for SVM, due to the different characteristics of the models.