Machine Learning Applications in Predicting Adult Education Dropout in Nigerian Non-Formal Education Systems

This study applies machine learning predictive modelling to adult education dropout prediction in Nigerian non-formal education systems, developing and validating a practical early warning system for programme managers and state education agencies. Adult education dropout is a persistent and costly problem in Nigeria's non-formal education system, but prediction and prevention efforts have relied on intuitive rather than data-driven approaches. Machine learning offers the potential to identify at-risk learners at early programme stages using routinely collected administrative and assessment data, enabling targeted retention interventions before dropout occurs. This study uses learner records from 18,000 adult education programme participants across five states, including enrolment demographics, attendance patterns, assessment performance, programme characteristics, and facilitator variables. Multiple machine learning algorithms are compared including logistic regression, random forest, gradient boosting, and neural network classifiers. Model performance is evaluated using area under the receiver operating characteristic curve, precision, recall, and F1 scores on held-out test samples. An explainability analysis using SHAP values identifies the most influential predictive features. Findings reveal that the gradient boosting classifier achieves the highest predictive accuracy with an AUC of 0.82 on the test set. The most predictive features are third-week attendance rate, initial assessment performance, facilitator experience level, and geographic distance from the centre. The study contributes an original machine learning dropout prediction model for Nigerian adult education and a practical early warning system prototype. It recommends piloting the system in UBEC-funded programmes in three states.

Keywords: machine learning, dropout prediction, adult education, Nigeria, early warning system.