A Machine Learning Approach for Data-Driven Decision Making in Student Academic Performance

Abstract

The early prediction of the student academic performance during a semester is a real challenge in educational data mining. When instructors know from the mid of the semester which students are at risk, target early interventions by adding tutorials, office-hour encouragement, remedial assignments can be implemented to improve outcomes. In this work, a complete machine-learning pipeline is applied to real assignments and exams data of 11 classes and 392 students in a general physics I course. In this study a Random Forest model was developed to predict the final total grades by using real data including quiz, homework and midterm grades. The regression model shows R2 of 0.823 and MAE of 6.43 points. The results were translated to a binary pass/fail classification, where the model reliably identifies risk students with strong Receiver Operating Characteristic (ROC) performance. The work provides a robust and scalable tool for early warning systems in higher education, supporting data-driven decision making and targeted academic interventions