On the Explanation of AI-Based Student Success Prediction

Abstract

Student success prediction is one of the many applications of artificial intelligence (AI) which helps educators identify the students requiring tailored support. The intelligent algorithms used for this task consider various factors to make accurate decisions. However, the decisions produced by these models often become ineffective due to lack of explainability and trust. To fill this gap, this paper employs several machine learning models on a real-world dataset to predict students’ learning outcomes from their social media usage. By leveraging the SHapley Additive exPlanations (SHAP) to investigate the model outcomes, we conduct a critical analysis of the model outcomes. We found several sensitive features were considered important by these models which can lead to questions of trust and fairness regarding the use of such features. Our findings were further evaluated by a real-world user study.