Predicting Diabetes Mellitus by Using Data Mining Techniques

Abstract

            This research aims to 1) develop a model for diabetic patient prognosis using data mining techniques and 2) compare the performance of the most suitable model for diabetic patient prognosis to create a prognostic model. Furthermore, compare the performance measures of data classification by the Accuracy (Accuracy) that gives the highest value. The sample used in this study was 300 items of patients who received screening for diabetes and 100 items of patients with diabetes, totaling 400 items. The research tools were questionnaires, and data were analyzed through the Waikato Environment for Knowledge Analysis program using multilayer perceptron, Naive Bay, logistics techniques Random Forest Technique, and decision tree technique.                           The research instruments tools include accuracy, precision, recall, F-measure, and ROC curve.

            The results showed that the Random Forest technique gave the highest accuracy in predicting the outcome of diabetes at 99.75% with a precision of 98.50%. The Recall was 98.50%. The overall efficiency measurement (F-Measure) was 98.50%, and the curve value (ROC) was 95.20%. The results of this research could be applied in treating diabetic patients in the future.