Features Importance to Improve Interpretability of Chronic Kidney Disease Early Diagnosis

Chronic Kidney Disease (CKD) is a worldwide chronic disease that, if it is recognized late, leads a vast majority of patients to suffer premature mortality and quality of life decline due to a progressive loss of kidney function. Data mining classifiers would contribute to an early diagnosis and hence preventing kidney severe damage since subtle patterns in CKD indicators can be discovered. By employing Cross Industry Standard Process of Data Mining (CRISP-DM) methodology along with features importance techniques, this work develops a classifier model that would support healthcare professionals in early diagnosis of CKD patients. By means of a data workflow pipeline, an automated data transformation, modelling and evaluation is applied to the CKD dataset extracted from the University of California Irvine-Machine Learning (UCI-ML) repository. The pipeline developed is used to carry out an exhaustive search of the best data mining classifier and the different parameters of the data preparation's sub-stages like data missing and feature selection. As a result, AdaBoost is selected as the best classifier with a 100% in terms of accuracy, precision, sensivity, specificity, and f1-score; outperforming the classification results obtained by the related works even with unseen data from a test set. Regarding model's interpretability, the application of feature selection reduces from 24 to 12 the group of features to be employed in the classifier model developed, achieving more explainable model's outputs. Furthermore, an analysis of the importance of features selected is carried out to explore the relevance of each selected feature.