Optimization-Based Hyperparameter Tuning Using Extra Trees to Classify Type-2-Diabetes Mellitus

The increasing prevalence of diabetes mellitus (DM) is impacting an ever-growing number of households globally. Many individuals diagnosed with diabetes have limited awareness of their health status and associated risk factors prior to diagnosis. This study introduces a novel approach utilizing Random Forest-based Grid Search Cross Validation (RF GSCV) for hyperparameter tuning, aimed at enhancing classification models to estimating type 2 diabetes (T2DM). The starting focus of this research is to improve the SOTA accuracy of the models and ensure their adaptability across diverse datasets. A dataset consisting of 520 individuals, characterized by various diabetic symptoms, was used to predict the likelihood of developing diabetes, considering both positive and negative probabilities. Optimal parameters were applied to three classification models—Random Forest (RF), XGBoost, and Extra Trees (ET)—to predict diabetes. The models is validated using 10-crossvalidation, with the Extra Trees model achieving the highest classification accuracy of 97.60%.