Predicting student performance in Massive Open Online Courses (MOOCs) using machine learning and deep learning techniques: a comprehensive survey

High dropout rates in online learning, particularly in Massive Open Online Courses (MOOCs), present a significant challenge to educational institutions. An accurate prediction of dropout risk is essential for designing effective intervention strategies. Unlike prior surveys, this comprehensive review synthesizes 57 studies from 2015 to 2024 to identify systematic methodological advances and critical limitations in dropout prediction. Our analysis categorizes the approaches across three dimensions: model type, learning paradigm, and performance characteristics. A key finding is the identification of inflection points: behavioral + temporal features improve accuracy by 15–25% over static features; ensemble methods achieve 85–92% accuracy with superior interpretability; and advanced Deep Learning (DL) architectures reach 95% accuracy but require substantial computational and data resources. These findings enable evidence-based model selection based on the institutional constraints. We critically evaluate the limitations of existing approaches, including inadequate class imbalance handling, lack of cross-platform generalization, and insufficient intervention validation, and propose standardized evaluation protocols. This survey serves as a practical reference for researchers and educators in designing data-driven dropout reduction strategies that align with institutional capabilities.