Demographics and other attributes such as learning approach, motivation, course perceptions and prior academic performance have proven to be effective predictors of academic performance in higher education. However, limited studies have explored the intricate relationships between these categories of predictors and their combined effect on academic performance. This study investigates the effect of these attributes on academic performance through the application of machine learning techniques, with a focus on decision tree algorithms. To enhance model generalisation and mitigate overfitting, K-fold cross-validation is implemented. In this approach, the dataset is partitioned into k equally sized folds; during each iteration, k–1 folds are utilized for training while the remaining fold serves as the validation set. Consistent with prior research, the findings highlight that prior academic performance is the most significant predictor of academic success. Additionally, learning approach emerged as a key factor, with better performing students having high scores in deep learning especially for younger female students with better prior academic performance. The study further revealed that prior knowledge significantly contributes to academic performance for a specific group of students, while age and gender also exert measurable effects on performance. Based on these insights, several recommendations such as bridging classes and course development with a focus on deep learning, were proposed to better support students in achieving academic success. In conclusion, this study provides valuable information to guide higher education institutions in enhancing pedagogy through a data-driven approach, ultimately fostering improved students’ academic performance.

Keywords: Demographics, higher education, academic performance, cross-validation, data mining.