This study examines the application of machine learning to predict academic performance in Massive Open Online Courses (MOOCs), aiming to address the persistent challenge of low completion rates despite high enrollment rates. Identifying at-risk students early in the course is crucial for enabling targeted interventions to improve engagement and retention. The research is essential because it offers a practical and scalable approach for early detection using only behavioral data readily available during the initial stages of learning. The study utilized data from a Coursera MOOC, comprising over 8,000 anonymized student records and more than 80 event log files. Aggregated behavioral features—such as total sessions, number of quizzes completed, lecture views, supplementary activity access, and average session duration—were extracted from the first 50% of course participation. Following the CRISP-ML(Q) framework, three predictive models were developed and compared: logistic regression, decision trees, and support vector machines. The results showed that the decision tree classifier achieved the highest performance, with 90.9% accuracy and balanced precision and recall. These findings suggest that early-stage, aggregated behavioral metrics can predict course outcomes without relying on complex sequence modeling. The implications are significant for MOOC providers and educational institutions seeking real-time, interpretable, and scalable solutions to support student success. The proposed model can guide early interventions and improve overall course completion rates in online learning environments.