Implications of Meta Classifiers for Onset Diabetes Prediction

Abstract

In the data mining area, the prophecy of human diseases initiates a research zone for researchers by applying various machine learning algorithms with various patterns. As a modern community disease, diabetes is becoming one of the fastest-progressive human diseases in the world because of eating heavily sugared foods and lack of proper diet knowledge. In this era, most of the middle age people have confusion about the presence of diabetes in their bodies. That’s why we choose to do research on diabetes. In this paper, we scrutinized the classification performance of six Meta Classifiers named as Multiclass Classifier Updatable, Attribute Selected Classifier, Ada Boost M1, Logit Boost, Bagging, and Filtered Classifier for forecasting diabetes through cross-validation and percentage split techniques using in WEKA whereas as a diabetes dataset we used Pima Indians Database. And finally, according to win-rate from the Win-Draw-Loss table, the highest performance comes from Multiclass Classifier Updatable which has an 80% win-rate. On the other hand, in the measurement of highest individual accuracy, 81.9923% comes from both Attribute Selected Classifier and Filtered Classifier. According to the measurement of the highest average performance, 66% Split as a percentage split technique and Attribute Selected Classifier show the highest performance.