Chronic Kidney Disease Detection: A Comparative study

Chronic Kidney Disease (CKD) is a life-threatening disease that impairs the kidney making it unable to filter blood properly. Early detection and appropriate treatment of CKD are crucial to increase the survival rate of patients. This study provides a comparative analysis for the identification of CKD. Toward this end, the performance of Logistic Regression, Random Forest, K-Nearest Neighbor (KNN), and Support Vector Machines (SVM) classifiers have been compared. The CKD dataset of the UCI machine learning repository was used to train the models. The dataset consists of 400 cases of which 246 (62%) have CKD and 149 (38%) are healthy cases. The dataset has 25 features distributed into 11 numerical features and 14 categorical features, including the response. The data were loaded into R for exploratory data analysis and pre-processing. To solve the issue of data imbalance, the researchers performed oversampling on the dataset. This study revealed that Random Forest outperforms others with an accuracy of 0.992 and 0.993 F1Score after oversampling. Additionally, the study found that oversampling increased the performance of Logistic Regression, Random Forest and SVM models. However, KNN experienced a decrease in performance after resampling, from 0.988 to 0.967 accuracy after oversampling.