Performance Comparison of XGBoost and Naïve Bayes Algorithms Using SMOTE-Tomek for Diabetes Disease Classification

Diabetes mellitus remains a major global public health issue, including in Indonesia, where the rising number of cases requires accurate and timely prediction. Classification models often face challenges due to imbalanced datasets, which reduce their ability to detect minority classes. This study aims to compare the performance of the XGBoost and Naïve Bayes algorithms and to evaluate the effectiveness of the SMOTE-Tomek technique in handling class imbalance for diabetes classification. The dataset used is the Kaggle Diabetes Prediction Dataset, consisting of approximately 100,000 records with demographic and clinical attributes. The preprocessing stage includes label encoding, Min-Max normalization, and data balancing using SMOTE-Tomek. The dataset is then split into training and testing sets in an 80:20 ratio, and model performance is evaluated using accuracy, precision, recall, and F1 Score. The results show that XGBoost outperforms Naïve Bayes, achieving 97% accuracy, 88% precision, 72% recall, and 79% F1-score. Furthermore, the application of SMOTE-Tomek improves detection of the minority class, particularly recall, while slightly decreasing precision and F1-score. The combination of XGBoost and SMOTE-Tomek proves more effective at handling imbalanced data for early diabetes detection.