An Ensemble Framework using Enhanced Gradient Descent and Butterfly Optimization for Chronic Kidney Disease Prediction

Chronic Kidney Disease (CKD) is a devastating medical condition that impacts millions worldwide and can pose serious health complications or even death if not identified early. Traditional machine learning models show some promise in CKD diagnosis, but they are challenged by overfitting, slow convergence, and local minima. This study introduces an enriched gradient-based descent that exploits a fusion of Butterfly Optimization Algorithm, called Enhanced Gradient-Based Descent with Butterfly Optimization (EGBDBOA) to overcome these limitations. Namely, the EGBD method allows for both a more efficient weight update utilizing an adaptive learning rate and BOA is used as a global search mechanism to prevent convergence into local optima. Combining the two method leads to more robustness and a higher level of accuracy with accelerated convergence speed. CKD dataset provided by UCI Machine Learning Repository. This dataset has patient demographics like, patient ID, Gender, clinical parameters, Blood Pressure, Specific Gravity and lab results (identity of RBC. Random Forest, model based on SVM multi model ACO ensemble predicted features, where thus achieving accuracy 97.25% and producing an $A U C=0.98$, and outperforming the baseline models such Random Forest, SVM, ACO). Novel Hybrid Model and its value proposition: This model reflects an innovative way of improving CKD prediction which can not only provide a new edge for health care providers to have more accurate predictions towards optimal treatment but also speed up the process.