Metode Bayesian dan Multilayer Percepton dalam Mengklasifikasi Diabetes Mellitus

Abstract

Diabetes mellitus is a chronic metabolic disorder that causes glucose regulation in the blood. Blood sugar anomalies can be defined as unwanted readings either due to normal causes or reasons unknown to the patient. Machine learning applications have been widely introduced in diabetes research and blood sugar anomaly detection. However, modeling options and strategies for classification in diabetes mellitus are needed. This study aims to classify the data as diabetic or non-diabetic and improve classification accuracy. Classification accuracy is improved by using many of the data sets as training data and test data. Classification accuracy is improved by using multiple of the data set as data. In the test, the C4.5 and RF hybrid methods, as well as the MLP and Net Bayes hybrid classification methods were developed for the classification of diabetes. In the case of C4.5 + RF it provides an accuracy of 79.31% which is higher than the individual models. Similarly, MLP + Net Bayes, provides an 81.89% higher accuracy than the individual models. In the second case the 85-15% ensemble model training and test partitions have an important role for the classification of diabetes data. The proposed MLP + Net Bayes provides 81.89% accuracy as a robust model for data classification. So that the proposed model achieves the highest accuracy of 81.89% with 6 features and reaches the highest sensitivity of 64.10% and the highest specificity of 90.90%.