<p>Handling imbalanced datasets is crucial for improving the efficiency of machine learning techniques in healthcare applications. In epilepsy datasets, seizure instances are often underrepresented compared to non-seizure instances, which negatively impacts the effectiveness of classification algorithms. This study investigates the use of SMOTE-ENN and SMOTE-Tomek sampling methods to address data imbalance and evaluates their effect on the performance of various state-of-the-art classifiers. A comparative analysis of minority class prediction revealed that SMOTE-Tomek achieved higher recall for Logistic Regression, KNN, SVM, SGD, and Random Forest, while SVM achieved the best F1-score. Perfect recall and F1-scores (1.0) were observed for Decision Tree, Extra Trees, Gradient Boosting, and XGBoost classifiers after applying both sampling techniques. Experiments were conducted on the standard public Bonn EEG dataset, using ten classifiers for binary seizure detection. Overall, Decision Tree, Gradient Boosting, and XGBoost achieved 100% accuracy, precision, recall, AUC and F1-score, demonstrating the effectiveness of data balancing. Results demonstrate that SMOTE-Tomek as well as SMOTE-ENN enhance minority class prediction, with SMOTE-Tomek providing superior performance for SVM and Random Forest across most metrics. These findings provide valuable insights for researchers to leverage data balancing techniques to improve classifier performance on imbalanced epilepsy datasets.</p>