Machine Learning-Based Predication of Chronic Kidney Disease

Kanchan Wavhale, Dr. Monika Rokade, Dr. Sunil Khatal

Chronic Kidney Disease (CKD) is a serious, progressive, and widely known medical condition that afflicts millions around the globe and often is not diagnosed until it has reached its later stages. Healthcare systems face obstacles in the timely diagnosis of CKD, due to the gradual nature of its progression and the lack of early warning symptoms. The older methods of diagnosis rely on clinical judgement alongside laboratory analyses, and this can result in hazardous delays in diagnosis. To this end, a new type of medical diagnostic tool that is faster than traditional methods and has the capability of pinpointing CKD in its earliest stages is necessary. The current paper will discuss machine learning algorithms as a novel means of diagnosing Chronic Kidney Disease and its detection in its earlier stages. Current machine learning frameworks will be discussed, including multi-class classifiers, and will conclude with a framework that provides promising results for the early detection of CKD. A significant portion of this framework will be dedicated to the issues of robust data pre-processing. The proposed methods will be discussed primarily in these terms, including the data pre-processing methods of imputation and normalization. These methods will also include the feature selection methods of ReliefF and Ranker, as well as selection of the most relevant classifiers, which will be described as a combination of artificial neural networks (ANN), J48 decision trees, naïve Bayes (NB), and logistic regression for the multi-class case. Techniques with the most promise for the enhancement of predictive performance and the reduction of classification error will also be discussed. The proposed methodology will be subjected to evaluation, with a 70% training - 30% testing approach to ascertain the rigor of the proposed measure. The experimental results indicate that the hybrid model has achieved a remarkable performance level of 96.67% in accuracy, 100% in precision, 96.67% in recall, and 98.31% in F1-score, which demonstrates that the proposed model is better than the single classifiers. Such results prove that the integration of diverse machine learning methods is beneficial for the model and enhances the improvement of the model in the diagnosis. The proposed system will enable healthcare practitioners to detect CKD at an early stage, which will allow them to manage patients in a better way, and in a timely manner. The research in general affirms the claim that hybrid machine learning models can be of significant value for improving the accuracy of medical diagnosis, and reinforcing the medical practitioners’ decision-making process in the clinical settings.