A Comparative Study of Gradient Boosting and Random Forest for Hyperglycaemia Prediction

In order to stop type 2 diabetes from developing and its related problems, hyperglycaemia must be identified early. Using clinical data that includes seven important characteristics (pregnancy, blood pressure, skin thickness, insulin, BMI, diabetes pedigree function and age), this study compares two ensemblebased machine learning models: Random Forest (RF) and Gradient Boosting (GB). Following preprocessing (imputation, scaling), both models were trained using stratified 1 0-fold crossvalidation after class imbalance was addressed with SMOTE. At 96.8% accuracy, Gradient Boosting outperformed Random Forest, which came in at 9 5. 2% accuracy analysis consistently found that the main predictors were age, BMI and pregnancy. These findings imply that the best model for early hyperglycaemia risk identification is gradient boosting, which may also offer useful assistance for clinical judgement.