Key points are not available for this paper at this time.
Student attrition significantly hinders the achievement of Sustainable Development Goal 4. School counsellors, crucial in ensuring students stay in school and graduate, often face challenges in analysing data to prioritise interventions due to the complexity and volume of data. This study explored using data mining to support counsellors in identifying at-risk students by focusing on non-academic variables. The Delphi technique was used to engineer new features. The dataset comprised 21 features and 598 records. Five classification algorithms were deployed in modelling, using tenfold cross-validation. Model performance was compared based on Accuracy, F1 Score, misclassification rate, Kappa statistic, ROC Area, and PRC Area. Random Forest emerged as the superior model, with a 96.7% precision and a weighted average Area Under the Curve of 0.996. Key predictors of attrition included high self-awareness, good social skills, high empathy, high emotional regulation, and being from a nuclear or single-parent family for low-risk. Average self-motivation and being part of extended or reconstituted families predicted moderate-risk. High-risk predictions were associated with low family income, lower parental education levels, low self-awareness, and neglectful parenting. These findings highlight the importance of incorporating non-academic factors in predictive models to better support at-risk students. Future research should explore real-time data integration, use larger datasets, and implement advanced class imbalance mitigation measures, to enhance model generalizability and effectiveness in student retention strategies.
Bakariwie et al. (Thu,) studied this question.