Detection of Chronic Kidney Disease Using Machine Learning Approach

The slow progression of chronic kidney disease (CKD) makes early detection and effective treatment the only ways to prevent the mortality rates. In this study, an amalgamation of ensemble machine learning (ML) models has been leveraged in an effort to support clinicians in their goal of faster, more accurate CKD recognition and detection. By detecting and assessing the risk variables early on, patients could limit the ramifications of this disease on their health. Consequently, binary categorization is the basis of this proposed ML technique. The CKD dataset, obtained from the UCI machine learning repository was utilized in this research consisting of 400 instances and 24 attributes, which is comprised of indicators, symptoms, and risk factors. 80% of the data was used to train the model, while the remaining 20% was used for testing. While utilizing the entire set of 25 features, the CatBoost and Random Forest models outperformed and outmatched the remaining algorithms with an accuracy of 99%. The Decision Tree, Ada Boost, and SVM algorithms were then used, with a constructive accuracy rate of 98%, 98%, and 95%, respectively. Furthermore, ROC curve for the five chosen ML models was used as a significant evaluation metric to help improve and supplement our understanding of the performance of the CKD categorization challenges. The results showed that the CatBoost model is more efficient and competent in successfully and accurately classifying a patient's CKD status, with an accuracy of 99.9% when critical attributes were used.