Quantitative Assessment of Explainability in Machine Learning Models : A Study on the OULA Dataset

Abstract

Many studies on AI in education compare model performance and fairness, but few focus on explainability. To address this gap, we evaluate two machine learning models—Artificial Neural Network (ANN) and Decision Tree (DT)—focusing on performance and explainability in predicting student performance using the OULA dataset. The DT, being inherently explainable, struggles with complex data relationships and misclassification, while ANN, although more accurate and stable, lacks transparency. Using the LIME method, the ANN outperforms the DT in accuracy and stability, but enhancing the interpretability of ANN models remains a key challenge for future research.

Ähnliche Arbeiten