A Comprehensive Analysis of Machine Learning Algorithms for Diabetes Prediction: Comparative Performance and Insights

Diabetes is one of the most chronic conditions seen worldwide. Its prevalence has been causing many complications, such as cardiovascular diseases, kidney failure, and neuropathy. Early prediction is crucial in helping prevent these from occurring or, at worst, to be affected minimally. In this work, we implemented various machine learning techniques to predict the onset of diabetes based on demographic and health-related data like age, body mass index (BMI), hypertension, heart disease, and blood glucose levels. The used algorithms were XGBoost, Random Forest, Stacked Ensemble, Gradient Boosting Machine (GBM), and Support Vector Machine (SVM). XGBoost could reach a maximum AUC of 0.979653 and a log loss of 0.0789284 with an F1-score of 0.811485. These really demonstrated strong power in the prediction of diabetes. High performance was also highlighted with the Stacked Ensemble Model, reaching an AUe of 0.979579 and an accuracy of 97.22%. These outcomes suggest that the superior ensemble methods, XGBoost and Stacked Ensembles, will provide quality prediction such that earlier diagnosis of patients with diabetes might be achieved consequently and, therefore, good recovery.