Comparative Analysis for Feature Extraction and Prediction of CKD Using Machine Learning

Abstract

Purpose: Chronic Kidney Disease (CKD) is one of the world’s top 20 causes of death. This novel study focuses on creating a prediction system for chronic kidney disease. It leverages T2 weighted MRI images and machine learning for efficient CKD classification, replacing labour-intensive manual processes. The adaptability of machine learning models accommodates changing disease patterns and diverse data sources. The purpose of this study is to investigate CKD, characterized by a sustained reduction in renal function lasting at least three months. CKD severity is gauged by kidney damage extent and glomerular filtration rate decline. The ultimate stage of CKD is end-stage renal disease. Methods: The study focuses on various feature extraction from MRI data using (kNN), Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT), and Gray level co-variance matrix (GLCM) along with a few morphological operations. Three Different sets of features are extracted, and Machine Learning Classification Models (Logistic Regression (LR), Support Vector Classifier (SVM), Decision Tree (DT), Random Forest (RF), k-Nearest-Neighbors (kNN), Naïve Bayes (NB)) are trained and tested on these set of features. Results: Experiment results show that LR Classifier gives the highest Accuracy of 92% for GLCM features. SVM and RF Classifier provide the highest Accuracy of 91.5% for DCT features, and RF Classifier gives the highest Accuracy of 86.6%. Based on predictions made by each model, a soft voting classifier is trained and tested to achieve the best Classification for each set of features. This study helps analyse the influence of the voting classifiers obtaining an accuracy of 90% for GLCM features, followed by 89% for DCT features and 84% for DWT features. Conclusion: The suggested system offers a comparative analysis of classification models and techniques for feature extraction. The results of performance evaluation illustrate the efficacy of different feature extraction and classification approaches. This will contribute to the timely detection, routine examination, and efficient control of CKD.