Predictive Modeling of Chronic Kidney Disease: An Ensemble ML Approach

Abstract

Globally, Chronic kidney disease (CKD) is becoming a significant threat to public health, As Effective management and treatment of CKD depend heavily on early detection. In this study, we propose an in depth approach for CKD detection through stacking of machine learning. We utilized a hospital dataset with 25 features to develop prediction models for the classification of chronic kidney disease. The dataset is intended for a classification challenge and contains multivariate data. After that, the data was divided into training and testing sets using an 80-20 split, making it possible to assess the performance of the model. Many machine learning models were used, however one stacked model that included the Random Forest Classifier, Gradient Boosting Machine (RG), Convolutional Neural Network (CNN), and Decision Tree received special attention. The suitability of these models for the CKD classification assignment led to their selection. After hyperparameter tuning was done to maximize the models' performance, the models were assessed using metrics including AUC, accuracy, F1 score, precision, and recall on the testing dataset. Our research showed that the layered RG model produced very good outcomes. Later, hyperparameter tuning was done to maximize the models' performance, the models were assessed using metrics including AUC, accuracy, F1 score, precision, and recall on the testing dataset. Our results showed that the layered RG model produced very good outcomes. These results demonstrate how machine learning can be used to diagnose chronic kidney disease (CKD) early and have positive effects on healthcare by providing a way to improve patient outcomes and healthcare management.