An Efficient Machine Learning Technique to Predict Chronic Kidney Disease (CKD)

Abstract

Chronic Kidney Disease (CKD) is a pressing global health issue that affects millions of individuals and requires early prediction to reduce severe complications and improve clinical outcomes. In this study, an efficient machine learning framework is proposed to predict CKD with high accuracy and generalization ability. Feature significance was analyzed using three statistical measures—heatmap correlation analysis, information gain, and standard deviation—calculated for average, minimum, and maximum values. To evaluate predictive performance, Logistic Regression (LR), Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP) classifiers were employed. Experimental findings reveal that the average feature scores deliver the highest classification accuracy (100%), while the minimum values yield reduced model complexity without sacrificing performance. Specifically, by eliminating nine non-critical features (specific gravity, packed cell volume, hemoglobin, red blood cell count, bacteria, coronary artery disease, pus cell clumps, anemia, and pedal edema), Logistic Regression achieved 100% accuracy. Likewise, the maximum-value-based evaluation reported an accuracy of 96.75% when one redundant feature was removed. These results demonstrate the effectiveness of selective feature elimination in minimizing computational load while maintaining robust prediction capability. The proposed methodology enhances early CKD detection, supports timely medical interventions, and offers a cost-efficient diagnostic tool. The novelty of this work lies in the integration of three feature importance measures to optimize model performance, thereby contributing a mathematically balanced and clinically significant framework for CKD prediction.