Predicting student next-term performance in degree programs using AI-based approach: a case study from Ghana

Student performance can fluctuate over time due to various factors (e.g. previous assignment grades, social life and economic conditions). Temporal dynamics, such as semester-to-semester variations and changes in students’ academic achievements, behaviors and engagement over time, can be critical factors in designing predictive models. It can be said that most existing work focuses on one-time forecasting of student performance in specific semesters, subjects or short online courses without considering temporal elements. In this paper, we present a student performance-based temporal dynamic approach to progressively predict semester-wise performance. Eight semesters of data representing 3,093 undergraduate Health Sciences students was collected from a public university in Ghana, analyzed, pre-processed and transformed into a time-series format. Then a dynamic experimental framework utilizing four different machine learning methods to predict student performance was created. This includes Random Forest, Support Vector Machine, Long Short-Term Memory and Bidirectional Long Short-Term Memory to predict student performance semester-wise over eight semesters. The results indicate that utilizing past students’ performance records obtained over time enhances the accuracy of forecasting their performance in future semesters. Moreover, the results evident that high school grades and semester GPAs are the most powerful discriminant features influencing the models’ performance, emphasizing the importance of consistent in-course performance.