Chronic Kidney Disease (CKD) is a long-term medical condition where the kidneys gradually lose their ability to function properly. CKD is affecting millions of people worldwide. Early detection of CKD is crucial for preventing severe complications and improving patient outcomes. Traditionally, CKD diagnosis has relied on manual analysis of clinical parameters and laboratory tests, which often lack scalability and precision. The advent of Artificial Intelligence (AI), which involves the use of algorithms and machine learning techniques, has revolutionized healthcare by enabling automated, accurate, and efficient disease detection and classification. Datasets play a pivotal role in developing AI-based diagnostic systems, as the quality and balance of data significantly influence model performance. The majority of existing research on CKD detection has focused on balanced datasets, where data samples are evenly distributed across classes, to recommend the most effective classifiers for detection. However, in real-world scenarios, datasets are often imbalanced, with minority classes underrepresented, leading to biased models and poor detection of critical cases. Therefore, adopting suitable techniques to handle these imbalances is necessary. In this view, this paper addresses this by evaluating the performance of various classifiers on both slightly imbalanced and severely imbalanced CKD datasets. Through comprehensive experimentation, the research identifies that RNN demonstrates robust performance across both slightly and severely imbalanced datasets by achieving 95.67% and 89.77% balanced accuracy, 91.8% and 80.55% Mathews Correlation Coefficient (MCC), 99.68% and 82.6% AUPRC, 46.667 and 36 LR+, 0.068 and 0.186 LR-, and 64.8833% and 89.0909% H – measure. This work highlights the importance of designing adaptable classification methods that cater to real-world dataset characteristics, thereby enhancing the reliability and applicability of AI-based diagnostic systems for CKD.
Reddy et al. (Sun,) studied this question.