Data to Diagnosis: Evaluating Machine Learning Algorithms for Predictive Healthcare in Diabetes

Diabetes mellitus, a chronic metabolic disease, presents alarming challenges to world health. It is vital to diagnose it early to prevent serious complications. In this research, eight machine learning algorithms—SVM, XGBoost, Naive Bayes, Logistic Regression, Gradient Boosting, KNN, Decision Tree, and Random Forest—are used on a formatted dataset with clinical and demographic attributes. Normalization and categorical encoding were done for preprocessing. Although no class-balancing methods (e.g., SMOTE or weighting) were used or hyperparameter tuning was performed, models were tested with accuracy, precision, recall, F1-score, and confusion matrices. Interestingly, the dataset is very imbalanced (~10% diabetic cases), and thus may influence sensitivity. Ensemble models, particularly Gradient Boosting and XGBoost, reported more than 91% accuracy. In spite of limitations, findings suggest the promise of ML in early prediction of diabetes.