Optimized decision tree classification method for diabetes prediction

Abstract

Diabetes is one of the most deadly chronic diseases because most sufferers do not realize they have it. A more accurate prediction of diabetes disease must be made to reduce the risk of bad things happening to sufferers. This research will optimize the decision tree (DT) classification method for diabetes prediction. Optimization is done by splitting criteria, splitting data, particle swarm optimization (PSO), and parameter optimization to find the highest and most accurate forecast of diabetes. Splitting criteria is done by comparing the results of three criteria, namely gain ratio (GR), information gain (IG), and gini index (GI). Splitting data is done by dividing training data and testing data into three comparison groups, namely 70:30, 80:20, and 90:10. The application of PSO and parameter optimization is carried out to increase the accuracy value. The processed data is taken from the UCI machine learning repository with 520 records and 17 attributes (1 class/label attribute). From the experiments, the GI criterion with splitting data 90:10 obtained the greatest accuracy of 98.08%, and the combination with PSO resulted in an accuracy of 97.66%. Meanwhile, parameter optimization with splitting data 90:10 combined with GR criteria resulted in the highest accuracy of 97.90%.