Classification Rules of Chronic Renal Failure from Diabetic Patients Using Data Mining Techniques

Abstract

This research aimed to construct and compare the performance of models for predicting chronic kidney failure in diabetic patients. Patient data was collected from the database of Phu Wiang Hospital, Phu Wiang District, Khon Kaen Province, from January 1, 2019, to December 31, 2021, totaling 3,225 records. The researchers selected predictive factors using the Gain Ratio and SMOTE methods. Then, the RIPPER technique, PART Decision List, Naïve Bayes, and FURIA were used to construct models to predict chronic kidney failure in diabetic patients. In this experiment, 10-Fold Cross Validation was used to cluster the data and evaluate the models. The performance was compared using accuracy, sensitivity, and specificity. The results revealed that FURIA, using data after factor selection and balance adjustment using the SMOTE method, exhibited the best performance with an accuracy of 94.11%, a sensitivity of 91.30%, and a specificity of 96.90%. Moreover, the model was able to generate rules for classifying chronic kidney failure. Most of the discovered rules were related to comorbidities, medication use, and factors such as body mass index, education level, and occupation, which were important factors in classifying patients at risk of developing chronic kidney failure from diabetes.