The application of Machine Learning (ML) techniques in Educational Data Mining (EDM) is crucial for determining student outcomes and implementing personalized, timely interventions. This study addresses the prediction of academic performance in Mathematics, a critical STEM subject, by conducting a rigorous comparative analysis between classification and regression machine learning approaches. This study investigates the performance of high-efficacy ensemble algorithms, specifically Random Forest (RF), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost) algorithms across both predictive tasks. Classification models have the ability to predict success or failure (a discrete binary outcome), while regression models predict the continuous numerical final score. The findings reveal that classification is excellent for initial risk identification, while regression approach, leverage the superior performance of ensemble methods like XGBoost, which offers greater predictive granularity and more actionable insights for personalized educational guidance. . In conclusion, it is, therefore, submitted that the performance of the Extreme Gradient Boosting algorithm provides a better classification accuracy of 0.795 and F1-Score of 0.795 when compared to the other classifiers, the Extreme Gradient Boosting classifier also achieved mean absolute error (MAE) of 6.95 and variance R² of 0.88 in regression this shows that the algorithm has strong predictive power for predicting continuous variables