Predicting Academic Dropouts Using AI and Behavioral Data: A Hybrid Deep Learning Framework for Student Retention

Environmental education is recognized as a cornerstone of sustainable development, aiming to cultivate ecological consciousness, critical thinking, and responsible citizenship among learners. However, student disengagement and dropout from sustainability-related programs pose significant challenges to institutions and global environmental objectives alike. Addressing these issues requires innovative solutions that blend pedagogy with technology to proactively identify and retain at-risk students. This research proposes a hybrid Artificial Intelligence (AI) framework for predicting academic dropout in environmental education using deep learning and explainable machine learning models. The approach integrates Convolutional Neural Networks (CNN) to extract behavioral features from student engagement data, Long Short-Term Memory (LSTM) networks to capture temporal academic patterns, and XGBoost to classify student risk based on structured academic, demographic, and participation features. To ensure ethical and transparent decision-making, Shapley Additive Explanations (SHAP) are incorporated, providing interpretable visualizations of the factors contributing to dropout predictions. The model was trained and validated using real-world LMS datasets reflecting diverse student behaviors across environmental education courses. Experimental results demonstrate that the hybrid model achieves an overall prediction accuracy of 94.2%, with a high early-warning recall of 90.3% for moderate-risk students—those who typically fall through the cracks in conventional systems. The fusion of behavioral, academic, and emotional cues enables the detection of latent disengagement patterns before they escalate into full dropout. More than a predictive tool, this framework supports institutional strategies for promoting environmental awareness, by ensuring continuity in ecological literacy programs and reinforcing student commitment to sustainability. It aligns directly with the objectives of the UN Sustainable Development Goals (SDG 4: Quality Education and SDG 13: Climate Action), providing both micro-level (individual learner support) and macro-level (policy and governance) benefits.This study contributes to the fields of educational data mining, environmental education, and AI for social good by presenting a scalable, interpretable, and high-impact model for dropout prediction. It offers a blueprint for how AI systems can be ethically embedded into sustainability curricula to drive engagement, equity, and long-term ecological stewardship.