<![CDATA[The application of machine learning models in digital education has become increasingly important for predicting student academic performance and supporting early academic interventions. However, predictive accuracy alone is insufficient if models introduce bias against certain demographic groups. This study presents a fairness analysis of student academic performance prediction using Random Forest and Support Vector Machine based on demographic attributes and digital learning activities. The Open University Learning Analytics Dataset (OULAD) comprising 32,593 student records was utilized, with academic outcomes reformulated as a binary classification problem (pass/fail). Model performance was evaluated using accuracy, precision, recall, and F1-score, while fairness was assessed using Demographic Parity and Equal Opportunity metrics on the sensitive attribute highest prior education level. Experimental results show that Random Forest achieved superior predictive performance (F1-score = 0.835) compared to SVM (F1-score = 0.746). From a fairness perspective, Random Forest demonstrated lower disparity, reducing demographic parity gap from 0.466 (SVM) to 0.380, and equal opportunity disparity from 0.391 to 0.199 across education-level groups., indicating an overall bias reduction of approximately 15–20% across education-level groups. The findings highlight that models with higher predictive accuracy do not necessarily ensure fairness across demographic groups. The novelty of this study lies in the integrated evaluation of demographic fairness and digital learning activities within the OULAD context, extending prior studies that focus primarily on performance optimization. This research underscores the importance of incorporating fairness-aware evaluation in educational predictive systems to support ethical, transparent, and equitable decision-making in digital education.]]>
