Accurate prediction of student academic performance is essential for universities seeking to improve learning outcomes and deliver timely, data-driven support. Prior work commonly uses regression to estimate Grade Point Average (GPA), yet numeric predictions can be difficult for administrators to translate into actionable risk levels. This study reframes the task as binary classification, categorizing students as good (GPA ≥ 3.00) or poor (GPA < 3.00) performers. Using 2,423 records from multiple programs at an Indonesian university, we combine academic indicators from the learning management system (login frequency, assignment submission, and forum activity) with socio-economic and digital behavioral variables (parental income, extracurricular participation, study-group involvement, and social media use). Seven machine learning models—Naïve Bayes, Generalized Linear Model, Logistic Regression, Deep Learning, Decision Tree, Random Forest, and Gradient Boosted Trees (GBT)—are benchmarked under a consistent evaluation design. Results indicate that integrating academic, socio-economic, and digital behavioral features improves classification performance, and ensemble methods outperform single, traditional models. GBT yields the best accuracy of 0.75, offering a practical basis for early-warning dashboards and targeted interventions. The study provides comparative evidence from Indonesian higher education and highlights the value of incorporating digital engagement signals alongside conventional academic data for more effective student support services.