Effective prediction of students’ academic performance is an important problem in educational data mining and has implications for early intervention and curriculum planning. This paper presents a stacking ensemble model, using Random Forest (RF) and Support Vector Machine (SVM) as the base learners, and Logistic Regression as the meta-learner, to classify the students into two categories: pass or fail. The data used contained a total of 395 students records, with 33 profiling students based on demographic, behavioral, and academic features. A large amount of data preprocessing was conducted, including label encoding, feature scaling, and defining the target based on final grades (G3 ≥ 10 = pass). The proposed stacking ensemble produced a test accuracy of 86.08%, demonstrating the overall expected success. The accuracy, recall, and F1-score for the pass class were 0.96, 0.83, and 0.89. For the fail class, they were 0.73, 0.92, and 0.81. The macro and weighted average F1-scores were both 0.86, showing the two classes have the same level of strength. The important features for success included prior academic performance, absenteeism, and parent's level of education as the most common predictors of student success. This ensemble performed better than individual classifiers and has potential for practical application in settings for early identification of academic risks. This study demonstrated the usefulness of hybrid machine learning implementations for predicting educational outcomes.