An Approach to Detect Chronic Kidney Disease (CKD) by Removing Noisy and Inconsistent Values of UCI Dataset

Abstract

AbstractWith the alarming rate of increase in chronic kidney disease (CKD) cases all over the world, researchers are trying to resolve it with state-of-the-art methods. It is evident that in a certain time period such disease gradually disrupts other organs functioning eventually causing death of patients. Early detection of CKD can diminish the chances of further damage to a great extent. Considering the UCI Machine Learning CKD dataset, this work attempts to present a more reliable approach, enabling handling of noisy data. However, CKD dataset contains noisy and inconsistent values, resulting in inaccurate prediction of CKD by using traditional machine learning algorithms. Therefore, this research presents an approach of handling noisy and inaccurate values of CKD dataset by employing a combination of deep neural network, statistical methods, Principal Component analysis (PCA), and “SMOTE”. Consequently, the refined CKD dataset coming out of the mentioned pre-processed methods is used in various machine learning methods. Our results showed that RF outperformed with 98.5% accuracy among Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Decision Tree (DT), Naive Bayes (NB), and Logistic Regression (LR) classifiers. Additionally, we found that the features such as serum-creatinine and blood urea exhibited their dominance in outcome prediction.KeywordsChronic kidney diseaseMachine learningPrincipal component analysisDeep learning imputation