Interpretable and ethical learning assessment transformer (IELAT): an explainable transformer model for personalized student assessments
S. Hariharasitaraman, Amudhavel Jayavel, G. Sambasivam, Shahab Saquib Sohail, Dag Øivind Madsen · Cogent Education · 2025
The intervention of Artificial Intelligence (AI) in education has introduced new possibilities for personalized learning, but it also brings challenges related to transparency, fairness, and adaptability. Traditional AI models in educational assessments often operate as ‘black boxes,’ leaving educators with limited understanding of the factors influencing predictions. This research introduces the Interpretable and Ethical Learning Assessment Transformer (IELAT), an innovative model that integrates a Transformer with Generalized Shapley Additive Explanations (G-SHAP) and attention mechanisms to address these issues. By explaining model decisions clearly, IELAT ensures transparency and trust while delivering personalized, adaptive assessments in real-time. The model was evaluated using the Open University Learning Analytics Dataset (OULAD), achieving 99.81% accuracy, 99.80% precision, and a 99.75% F1 score. IELAT surpasses other models in improving educational outcomes by delivering ethical, data-driven assessments that ensure transparency, fairness, and adaptability in AI-driven education.