Performance evaluation of supervised machine learning algorithms for diabetes prediction

This study presents a performance evaluation of six machine learning algorithms for the prediction of diabetes using a publicly available dataset from Kaggle, which includes relevant clinical and demographic features. Comprehensive preprocessing procedures were undertaken to address missing values, outliers, and a pronounced class imbalance (38:1 ratio of non-diabetic to diabetic cases), which poses significant challenges to model performance. The evaluated algorithms—Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), Naive Bayes, and Artificial Neural Network (ANN)—were assessed using key performance metrics, including balanced accuracy, precision, recall, F1-score, sensitivity, specificity, and detection rate. Among the models, Random Forest achieved the highest balanced accuracy (93%), followed by SVM (83%) and Decision Tree (82%), demonstrating superior predictive performance. The findings underscore the potential of machine learning in enhancing diabetes diagnosis and management and conclude that a careful balance of performance metrics, especially when dealing with imbalanced healthcare datasets should guide model selection.