Assessment and Recurrence of Kidney Stones Through Optimized Machine Learning Tree Classifiers Using Dietary Water Quality Parameters and Patient’s History
Kidney stone disease is a result of combination of food items consuming, drinking water quality and genetic heritability, which has been observed to be more prone (both occurrence and recurrence) to certain geographic regimes as Thanjavur suburbs of Tamil Nadu in southern India. The research carried out involves collection of medical information of Kidneystone patients of the study area and survey of their dietary habits including drinking water quality (through laboratory study), selection of suitable classifier to model the Kidney stone recurrence with the most contributing of 22 parameters (with due model evaluation). Weka (3.8.1) machine learning framework was used for the study, for evaluating the model accuracy of 66 classifiers, resulting 22 classifiers with accuracy higher than ZeroR, which was considered to be the benchmark. Based on this study, C-4.5 classifier (called J48 in Weka) was found to be most robust classifier, based on accuracy, precision, Recall, F-Measure, MCC, ROC Area and PRC Area. The selected classifiers were again evaluated based on domain conformance (namely, literature, logic and consistency) to obtain four validated classifiers, thereby providing seven parameters and their threshold value for kidney stone recurrence, namely, family history (Yes), Sulphate (>17ppm), potassium (>74 ppm), nitrate (>1.2 ppm), salinity (>120 ppm), conductivity (<=289 ppm) and water consumption (moderate).