Optimizing Random Forest for Type II Diabetes Prediction Using Firefly Algorithm

This study develops an enhanced Random Forest (RF) model for Type II diabetes prediction, optimized using the Firefly Algorithm (FA) for feature selection. The aim was to address the limitations of standalone RF models, which often suffer from suboptimal performance without intelligent feature selection and parameter tuning. A standardized preprocessing pipeline using StandardScaler was implemented on a clinical dataset. The FA was then employed to identify an optimal feature subset, which was used to train the RF classifier. The hybrid FA-RF model achieved a superior accuracy of 99.91%, along with a sensitivity of 98.96%, specificity of 100%, and precision of 99.95%, significantly outperforming the baseline RF model (97.06% accuracy, 69.15% sensitivity). Results demonstrate that integrating metaheuristic optimization with ensemble learning creates a more accurate and reliable tool for early diabetes prediction. The model's high performance and robustness highlight its strong potential for deployment in real-world clinical settings as a decision-support system to improve screening and patient outcomes.