Comparative Analysis of Machine Learning Models for Diabetes Prediction

Early detection of diabetes can reduce serious complications. This paper compares eight supervised machine learning classifiers—Logistic Regression, k-Nearest Neighbors, Support Vector Machine, Naive Bayes, Decision Tree, Random Forest, AdaBoost and Gradient Boosting—for diabetes prediction on the Pima Indians Diabetes dataset (768 records, 8 features). All models share the same preprocessing and are evaluated with a hold-out test set, 10-fold cross-validation repeated three times, hyperparameter tuning, and class weighting, using accuracy, precision, recall, F1-score and ROC-AUC. Under repeated cross-validation, Gradient Boosting had the highest accuracy (76.9%) and F1-score (0.649), while Logistic Regression had the highest ROC-AUC (0.837). The Decision Tree detected the most diabetic cases on the test set (72.2% recall). Class weighting raised the recall of Logistic Regression from 50.0% to 70.4%. Differences between the top models were within one standard deviation, and glucose was the most influential feature.