Bio-Inspired Optimization for Improved Estimation of Glomerular Filtration Rate in Patients With Diabetes

Chronic kidney disease (CKD) persists as a major global health concern, emphasizing the need for precise and effective approaches for its early detection and management. This research presents a comparison of different approaches for predicting the estimated Glomerular Filtration Rate (eGFR) serves as a vital marker of kidney function. The study evaluates the performance of a standard XGBoost regression model and an optimized XGBoost model using Particle Swarm Optimization (PSO) for hyperparameter tuning. The CKD-EPI equation, a traditional clinical method, serves as a baseline for comparison. The focus is on improving prediction accuracy through the application of machine learning (ML) techniques, particularly with the optimization of model parameters using bio-inspired algorithms like PSO. Evaluation metrics include Mean Squared Error (MSE) and R-squared ($\mathbf{R}^{\mathbf{2}}$). Results show that the PSO-optimized XGBoost model outperforms both the CKD-EPI equation and the default XGBoost model, demonstrating the effectiveness of machine learning and bio-inspired optimization techniques in enhancing CKD prediction accuracy. The PSO-optimized XGBoost resulted in a MSE of 808.35 and an $\mathbf{R}^{\mathbf{2}}$ score of 0.0086, demonstrating its potential for enhanced eGFR prediction.