Enhancing Diabetes Prediction: A comparative analysis of Train-Test Split and Stratified 10-Fold Cross-Validation with SMOTE Integration

Diabetes is a metabolic disorder identified in a human being by an increase in blood sugar levels, which poses a significant global health challenge. As the risk of diabetes is continuously increasing, it is crucial to predict diabetes in earlier stages. This study explores the use of two different data-splitting techniques employing five different ML algorithms, Train-Test Split and Stratified 10-fold cv. The main aim of this study is to improve model accuracy and to learn about the different splitting techniques present. The Random Forest Model attains an accuracy of 91.78% in train test split and 86.18% in Stratified 10-fold cv, which comes out to be the best model in terms of accuracy in both splitting techniques. Moreover, to remove any bias present in the model or to improve models’ performance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied post the initial developments. Comparatively, Random Forest and XGBoost consistently demonstrate superior performance compared to other ML models employed in this study.