Customer churn prediction in the telecommunications industry remains a critical challenge, particularly due to inherent class imbalance in customer datasets. This study presents a comprehensive analysis of the performance of various machine learning algorithms combined with different sampling techniques for predicting customer churn in a Telco dataset. Six classification algorithms, like random forest, decision tree, Support Vector Machine (SVM), K-nearest neighbors, logistic regression, and gradient boosting, were evaluated using oversampling (SMOTE, ADASYN), undersampling (ENN, Tomek), and hybrid (SMOTE-ENN) methods. Performance was assessed using multiple metrics including precision, recall, F1-score, accuracy, AUC, and specificity. The results demonstrate that decision tree with Tomek Links undersampling achieved the highest overall performance (F1-score: 0.75, AUC: 0.87, accuracy: 0.99), followed closely by logistic regression with ENN undersampling. Notably, undersampling techniques consistently outperformed oversampling methods across all algorithms, while the hybrid SMOTE-ENN approach showed significant advantages and close to the best result of oversampling/undersampling. This study provides valuable insights for practitioners in selecting optimal algorithm-sampling combinations for telco churn prediction and contributes to the broader understanding of handling class imbalance in customer analytics.