COMPARISON OF DECISION TREE, NAÏVE BAYES, AND NEURAL NETWORK ALGORITHM FOR EARLY DETECTION OF DIABETES

Abstract

Diabetes mellitus is included in the top 3 most deadly diseases in Indonesia. Based on WHO data in 2013, diabetes contributed 6.5% to the death of the Indonesian population. Diabetes is a chronic disease characterized by high blood sugar (glucose) levels that exceed normal limits. In the health sector, historical medical data can be processed to extract new information and can be used for decision-making processes such as disease prediction. This study aims to classify predictions for early detection of diabetes in order to obtain accurate results for decision making. The data used are historical data on hospital disease patients in Sylhet, Bangladesh in the form of a diabetes dataset from the UCI Repository. The algorithms used are Decision Tree, Naive Bayes, and Neural Network. Then the three methods are compared using the Rapidminer tools. The measurement results are 90% accuracy with Decision Tree, 80% with Naive Bayes, and 70% with Neural Network. So that the best algorithm is obtained, namely the Decision Tree for predicting early detection of diabetes. Rule in the form of a decision tree generated from the Decision Tree is used for input or ideas for decision making in the health sector for diabetes.