Integrating Feature Selection and Extraction with Tuned Multilayer Perceptrons for Predicting Student Dropout Risk in Higher Education

Every year, thousands of students drop out of educational institutions, significantly undermining the effectiveness of our education systems. Identifying factors contributing to dropout and implementing early interventions based on these factors is crucial. However, the acquisition of institutional data poses significant challenges due to privacy and security concerns. Moreover, while the availability of educational datasets is increasing, these often suffer from poor structure or organisation or lack relevance to specific problems. Additionally, the tools commonly used in pedagogy may lack optimised processes. In this paper, we have proposed a novel approach to predict dropout-related factors based on socio-economic and academic features. Our method combines Random Forest for feature selection and Principal Component Analysis for feature extraction, followed by training a finely tuned neural network to classify potential dropouts effectively. We have evaluated our model on two benchmark datasets SATDAP and ICFES, achieving accuracies of 90.08% and 80.03%, respectively, thereby outperforming state-of - the-art machine learning classifiers. This demonstrates the potential of our approach in addressing the dropout challenge in educational settings.