Enhancing Chronic Kidney Disease Prediction with Hybrid Machine Learning Approaches

Abstract

ABSTRACT

Chronic Kidney Disease (CKD) is a growing global health concern that demands timely and accurate diagnosis to prevent severe health deterioration and reduce patient mortality. Traditional diagnostic methods often rely on manual assessments and may be constrained by the availability of expert medical personnel. To address this challenge, we propose a hybrid machine learning framework that combines the dimensionality reduction capability of an autoencoder with the robust classification power of a Random Forest ensemble model. The raw clinical dataset, sourced from the UCI Machine Learning Repository, undergoes thorough preprocessing, including missing value handling, feature normalization, and categorical encoding. The processed data is then passed through an autoencoder, where the encoder compresses the high-dimensional input into a latent space that captures the most informative patterns. These compressed features are subsequently used to train the Random Forest classifier for binary classification predicting whether a patient is affected by CKD or not. The proposed model achieved an accuracy of 99.8%, significantly outperforming traditional approaches and showcasing high potential for early-stage CKD diagnosis. This hybrid approach enhances prediction accuracy, reduces overfitting, and offers a reliable solution for integration into real-world clinical decision support systems, particularly in scenarios with limited access to specialized healthcare providers.