Virtual classrooms have become essential in higher education due to their flexibility and accessibility, but they also present challenges in monitoring student engagement and predicting academic success. This study proposes a hybrid data mining framework to analyze student behavior by integrating clustering, association rule mining, and classification techniques. We analyzed over four million interaction records from the Moodle platform of the online Psychology program at the Technical University of Manabí. After data cleaning and transformation, clustering revealed four distinct behavioral profiles, from highly engaged and high-performing students to those with low interaction and academic risk. Association rules identified key activities linked to better performance, such as forum participation and timely assignment submissions. Classification models, particularly Random Forest and Support Vector Machines (SVM), achieved up to 91% accuracy in predicting student performance. The proposed framework allows early detection of at-risk students and offers valuable insights into behavioral patterns in virtual learning environments. These results contribute to the development of effective, data-driven strategies aimed at improving student retention and academic outcomes in online education.