Integrating Psychological Stress Indicators with Academic Data for Student Dropout Prediction: A Decision Tree and Expert System Approach

Abstract

Student attrition remains a major problem in higher education. Although academic variables are well-established moderators, psychological wellness, especially stress, is an important but often ignored moderator. The purpose of this study is to construct prediction models for students at risk of dropping out by combining academic and psychological information. One major challenge in this field is the class imbalance of student records, which results in a significant drop in the dropout rate compared to the general population. Therefore, in this study, we employ a Decision Tree algorithm and use a Forward Chaining inference engine along with the Synthetic Minority Over-sampling Technique (SMOTE) to solve it. We employed a data set of 122 students at one institution, with psychological stress scores generated from a standardised questionnaire according to well-known symptom domains. The accuracy for the model with only a Decision Tree was 95.83%. For the stress score, integration with the FC-based attribute increased performance to 96.67%; however, this model exhibited only marginal improvement over the final model due to its very low accuracy when compared to that of SMOTE. This ensemble model performed the best with an accuracy of 97.50% and an AUC of 96.35%. This progression demonstrates that even though the introduction of psychological information is beneficial, an approach to balance data and ensure a robust prediction system is required. This article is a proof-of-concept analysis which creates an opportunity for universities to establish proactive, early-warning-driven models; yet there is a requirement for future validation studies on larger and more diversified samples.