A Novel Ensemble Machine Learning Framework for Improved Diabetes Prediction and Complication Prevention

This study introduces an innovative framework for predicting diabetes, employing advanced imputation strategies and ensemble machine learning techniques to boost prediction accuracy. Given the irreversible nature of diabetes, early diagnosis is paramount. We applied machine learning algorithms on the Pima Indian Diabetes (PID) dataset, comprising data from 768 patients with nine distinctive attributes, sourced from the UCI Machine Learning Repository. Our methodology integrates KNN Imputer for addressing missing data, SMOTE to tackle class imbalance, and RFE for optimal feature selection. We tested six machine learning algorithms: LR, DT, RF, SVM, Gradient Boosting (GB), and an ensemble Voting classifier. The ensemble model, which combines LR, DT, RF, SVM, and GB, outperformed individual models, achieving a top accuracy of 84.2%, with a root mean squared error (RMSE) of 0.334 and an R² score of 0.528. This study highlights the enhanced predictive power achievable through the synergy of multiple machine learning models.