LightWeight Probabilistic Ensemble Model for Chronic Kidney Disease Detection

Abstract

CKD is the prolonged disease caused by gradual damage and reduction in the function of kidneys. The symptoms develop slowly and are non-specific to the disease. It is marked by high morbidity, mostly leading to the development of other serious health issues or mortality. Due to its asymptomatic nature, it is rarely diagnosed at early stages until it has progressed. Henceforth, early detection of CKD is crucial as there is more possibility of reducing the progression of the illness and improving treatment outcomes, once it is detected at an early stage. This work proposes a LightWeight Probabilistic Ensemble model that uses Logistic Regression and Naïve Bayes (LWPE-LRNB) with soft voting strategy to identify CKD with more accuracy and effectiveness. Several Machine Learning (ML) algorithms namely Decision Tree (DT), Logistic Regression (LR), KSTAR, Support Vector Machine (SVM) and Naïve Bayes (NB) are also compared with the proposed system. The metrics considered for evaluating the performance of ML models are Accuracy, Precision, Recall and F1-measure. Based on the performance values, it is inferred that our proposed system achieved an accuracy of 91.5%, a precision of 0.95, a recall of 0.92 and an F-measure of 0.93. Compared to other ML models, our lightweight probabilistic ensemble model implies the highest performance to ascertain CKD symptoms as the hybrid probabilistic model captures both the linear relationship and feature distribution. With its enhanced performance, it is concluded that our system would be the highly promising tool to the early detection of CKD which in turn supports timely diagnosis and treatment to improve patient outcomes.