A Machine Learning-Based Predictive Framework for Proactive Early Academic Intervention in Higher Education

Abstract

Predicting student academic performance has become a concrete challenge for higher education institutions trying to keep students from slipping through the cracks. The earlier you catch a struggling student, the more you can do about it—yet most traditional assessment systems flag problems only after the damage is done. This paper presents the AI Student Performance Predictor. It is a machine learning system that classifies student performance into three categories: High, Medium, and Low. The system runs a Random Forest Classification algorithm trained on a synthetic dataset of 250 student records built around five features: Cumulative Grade Point Average across three semesters, internal exam marks, class attendance, daily self-study hours, and weekly practice questions. A weighted scoring method combines these features to reflect their relationship to academic outcomes. The Random Forest model achieved 82.0% classification accuracy on the test set, outperforming five other classifiers, including Decision Trees (74.0%), Support Vector Machines (76.0%), and Logistic Regression (70.0%). Feature importance analysis showed CGPA as the strongest single predictor (31.2%), followed by internal marks (24.8%), attendance (19.7%), study hours (14.3%), and practice questions (10.0%).