Predicting Dropout Risk in Adult Online Learning Using Deep Learning Model

Abstract

The rapid expansion of online education has created new opportunities for adult learners, offering flexibility and accessibility beyond traditional classrooms. However, persistently high dropout rates often exceeding 70% remain a critical challenge, undermining the effectiveness and sustainability of online learning. This study proposes a deep learning-based ensemble model to predict dropout risk among adult learners in online environments. Unlike traditional statistical or single-model machine learning approaches, the ensemble framework integrates multiple deep learners, including Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and Transformer-based architectures, combined through majority voting to enhance predictive accuracy and robustness. The research employs a quantitative design, using both primary data from the National Open University of Nigeria and secondary data from Kaggle repositories, encompassing demographic, behavioral, engagement, and performance-related features. Through systematic phases of data collection, preprocessing, model training, and evaluation, the framework aims to identify key dropout factors, ensure generalizability across diverse learner populations, and improve interpretability for actionable insights. Preliminary results demonstrate that ensemble deep learning outperforms traditional methods in handling imbalanced datasets and capturing complex learner behaviors. The findings contribute to the development of early warning systems that enable timely interventions, ultimately enhancing retention and supporting the success of adult learners in online education.