A Bayesian Optimization Framework for the Prediction of Diabetes Mellitus

Abstract

The advances of bioinformatics and medical sciences have generated an enormous amount of data which can be used by machine learning (ML) and data mining (DT) methods to transform the data into valuable knowledge and can improve diagnosis, prediction, and management of most chronic diseases. One of the most life-threatening and widespread chronic diseases is Type 2 Diabetes Mellitus (T2DM), characterized by impaired operation of glucose homeostasis. We used several cutting-edge machine learning algorithms including Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), Naive Bayes (NB) on diabetes data. A state-of-the-art Bayesian Optimization (BO) has been proposed to optimize the hyper-parameters of machine learning classifiers for the Diabetes Mellitus (DM). The optimized hyperparameters using BO achieved an accuracy of 77.60% with RF, 76.04% with SVM, 71.61% for DT, 73.96% for NB classifier. We also achieved 64.06% accuracy without BO optimized SVM. We justified our models using confusion matrix for each classifier. The statistical comparison among different classifier’s performances has been presented using the Boxplot and Analysis of variance (ANOVA) test.