A Hybrid Discriminative–Generative Ensemble Learning Framework for Early Prediction of Chronic Kidney Disease Using Machine Learning

Abstract

Worldwide, chronic kidney disease (CKD) happens to be one of the topmost reasons for mortality and morbidity due to the progressive and irreversible nature of the disease. Identification at an early stage is critical in improving the outcome of patients while reducing the risk of disease progression. Though promising findings have been obtained regarding the use of machine learning in the diagnosis of chronic kidney disease (CKD), most of the literature available is concentrated either on single classifier or on ensemble learning methods, with little emphasis on the combination of multiple learning approaches in a single predictive approach. The goal of the study is to predict chronic kidney disease (CKD) using Logistic Regression, Support Vector Machine, Multi-Layer Perceptron, Random Forest, and Gaussian Naive Bayes through the Hybrid Discriminative and Generative Ensemble Learning Framework. The proposed method improves the prediction capabilities of the system by incorporating the steps of preprocessing, transformation of features, hyper-parameter optimization and ensemble learning through Stacking, Hard Voting, and Soft Voting. The proposed system was evaluated using the publicly available CKD dataset with 389 patients. The experimental results show that the proposed hybrid framework produces a near-optimal predictive performance in comparison to the other standalone baseline models. Evaluation metrics tend towards the optimal value in a number of ensemble implementations. The results reveal the prospects of the proposed architecture in terms of being used as an intelligent decision support system in the context of early CKD assessment.