Integrating Random Forest with Neutrosophic Logic for Predicting Student Academic Performance and Assessing Prediction Confidence

Abstract

This study proposes a hybrid approach to predict students’ final academic performance in a mathematics course by integrating Random Forest, a supervised machine learning model, with neutrosophic logic to assess prediction reliability. The objective is to improve educational forecasting by not only predicting grades but also quantifying the confidence of each prediction through neutrosophic components—truth (T), indeterminacy (I), and falsity (F). The model was trained on a dataset of demographic, academic, and social attributes from Portuguese schools, achieving robust performance (MAE = 1.54, R² = 0.61). Key contributions include: (1) a framework for transparent AI-assisted decision-making in education, (2) actionable insights for identifying at-risk students, and (3) a novel application of neutrosophic logic to interpret prediction uncertainties. The results demonstrate the potential of combining machine learning with neutrosophic analysis to improve academic interventions.