Performance Evaluation of Machine Learning Algorithm Using Chronic Kidney Disease (CKD) Dataset

Abstract

Machine learning (ML) algorithms enable computers to recognize patterns and make predictions or decisions from data, rather than relying on explicit programming. This paper presents a predictive model for the early detection of CKD through ML. The study uses five years of Electronic Health Record (EHR) data from a diverse patient group. The dataset contains demographics, clinical history, lab results, medication information, and diagnostic codes. The research starts with 25 variables, in addition to the class property, and then reduces this to 15 by using Principal Component Analysis (PCA). This aims to reduce the number of parameters to find the best subset for identifying CKD. The research uses common ML algorithms—Support Vector Machines (SVM), Random Forest, and Logistic Regression—and assesses their ability to detect CKD early. When comparing the classification algorithms, Random Forest (RF) had the best accuracy, at 81.2967%.