Influence of Class Imbalance and Resampling on Classification Accuracy of Chronic Kidney Disease Detection

Abstract

Chronic kidney disease is one of the leading causes of death around the world. Early detection of chronic kidney disease is crucial to the reduction of mortality caused as a result of the disease. Machine learning methods are recently becoming popular for the detection of chronic kidney disease. This study investigates the influence of resampling for chronic kidney disease detection using an imbalanced chronic kidney disease dataset. Choosing an optimal feature subset for medical datasets is important for improving the performance of data-driven predictive models. The influence of imbalanced class distribution on predictive models has become an increasingly important topic due to the recent advances in automatic decision-making processes and the continuous expansion in the volume of the data collected by medical institutions. To address the identified research gap, an experimental evaluation of synthetic minority oversampling and near miss undersampling technique was performed on a real-world chronic kidney disease dataset using several classification methods such as decision tree, random forest, K-nearest neighbor, adaptive boosting, and support vector machine. The results demonstrate that a number of variables, including performance metrics, classification algorithm, and dataset characteristics, influence the best class distribution.The study also offers useful information about resampling methods for an imbalanced classification problem which will help improve classification accuracy.