SMART DROPOUT RISK PREDICTION USING STACKED ENSEMBLE LEARNING ON MULTI-MODAL EDUCATIONAL DATA

Abstract

Learning fosters opportunity, critical skill development, and breaks cycles of poverty, playing a vital role in personal and societal growth. Academic excellence and reducing student dropout rates remain key challenges. Dropout is a chronic issue that affects student achievement and institutional effectiveness. Existing prediction models often rely only on structured data such as grades and attendance, ignoring unstructured inputs like behavior, emotional feedback, and peer interactions. This study proposes a smart dropout risk prediction method using a stacked collaborative learning architecture that integrates multi-modal educational data. By combining structured information (e.g., academic performance and punctuality) with unstructured data (e.g., student feedback and social interactions), the model aims to improve accuracy and early identification of at-risk students. The process includes data pre-processing, multi-source integration, feature construction, and training base learners such as Gradient Boosting, Random Forest, SVM and k-NN. The system also addresses key real-world issues like adaptability, accessibility, and data imbalance. Experimental results on institutional and public datasets show the proposed model outperforms conventional approaches, achieving over 92.8% precision, 94.12% accuracy, and a reduced false-negative rate. This intelligent system supports proactive interventions to improve student retention and academic outcomes.