Handling Imbalance Data in Random Forest Classification Models for Diabetes Prediction

Diabetes, a chronic disease marked by elevated blood glucose levels due to insufficient or ineffective insulin production, is a growing global health concern. Its prevalence has risen from 4.7% in 1980 to 8.5% in 2014, with an estimated 700 million cases by 2045, primarily in developing countries. While there is no cure, early and accurate prediction using technology can aid prevention and management. This study evaluates resampling techniques to improve the performance of the Random Forest model in detecting diabetes on imbalanced datasets. Techniques tested include the model without resampling, BorderlineSMOTE, SMOTEENN, and Condensed Nearest Neighbour, assessed using accuracy, precision, recall, and F1 score. BorderlineSMOTE achieved the highest accuracy (98.80%) and optimal precision-recall balance, making it the most effective method. SMOTEENN delivered perfect precision (100%) but lower recall, while Condensed Nearest Neighbour showed minimal improvement. BorderlineSMOTE proves the best approach for addressing data imbalance and enhancing diabetes detection with Random Forest.