The rapid expansion of online learning platforms, particularly Massive Open Online Courses (MOOCs), has revolutionized global access to education. However, these platforms continue to face a significant challenge - high student dropout rates. Early identification of atrisk learners is essential for implementing timely interventions and improving course completion outcomes. This study proposes an intelligent deep learningbased framework to predict student dropout in online courses by analyzing behavioral and interaction data. A Long Short-Term Memory (LSTM) neural network is employed to model temporal patterns in learner activities, including video views, quiz attempts, and forum participation. Using the KDD Cup 2015 XuetangX dataset, the proposed model demonstrates high accuracy in predicting dropout risk within the initial weeks of course engagement. Comparative evaluation with traditional machine learning models reveals that the LSTMbased approach more effectively captures sequential engagement trends and identifies critical behavioral indicators influencing dropout. The findings provide a scalable and interpretable solution for online education providers to enhance learner retention, support personalized learning, and strengthen data-driven decision-making in digital education environments.
Dropout Prediction, Online Learning, Deep Learning, MOOC Analytics, LSTM, Student Engagement, Educational Data Mining, Early Intervention, Learning Analytics, Behavioral Time Series