A Robust Chronic Kidney Disease Classifier Using Machine Learning

Clinical support systems are affected by the issue of high variance in terms of chronic disorder prognosis. This uncertainty is one of the principal causes for the demise of large populations around the world suffering from some fatal diseases such as chronic kidney disease (CKD). Due to this reason, the diagnosis of this disease is of great concern for healthcare systems. In such a case, machine learning can be used as an effective tool to reduce the randomness in clinical decision making. Conventional methods for the detection of chronic kidney disease are not always accurate because of their high degree of dependency on several sets of biological attributes. Machine learning is the process of training a machine using a vast collection of historical data for the purpose of intelligent classification. This work aims at developing a machine-learning model that can use a publicly available data to forecast the occurrence of chronic kidney disease. A set of data preprocessing steps were performed on this dataset in order to construct a generic model. This set of steps includes the appropriate imputation of missing data points, along with the balancing of data using the SMOTE algorithm and the scaling of the features. A statistical technique, namely, the chi-squared test, is used for the extraction of the least-required set of adequate and highly correlated features to the output. For the model training, a stack of supervised-learning techniques is used for the development of a robust machine-learning model. Out of all the applied learning techniques, support vector machine (SVM) and random forest (RF) achieved the lowest false-negative rates and test accuracy, equal to 99.33% and 98.67%, respectively. However, SVM achieved better results than RF did when validated with 10-fold cross-validation.