MACHINE LEARNING ENSEMBLE CLASSIFIERS TO PREDICT TYPE 2 DIABETES IN A COHORT STUDY

Diabetes, being a chronic disease, has the potential to result in various severe health complications, such as stroke, heart attack, chronic kidney diseases, and other associated ailments. The main objective of this research was to develop the ensemble ML-based learners for predicting diabetes. Logistic regression (LR), ML-based classifiers, decision tree (DT), random forest (RF), bagging classifier (BC), boosting classifier (AdaBoost), gradient boosting decision tree (GBDT), and supporting vector machine (SVM) were utilized for detecting the diabetic cases and determination of most important attributes related to Type 2 diabetes mellitus (T2DM). The performance of the methods was evaluated by 5-fold validation method through reporting accuracy, sensitivity, precision, the area under curve (AUC), and other indices. The Shahedieh cohort dataset in Yazd province including 9398 participants, conducted from 2014 to 2016, got used in this study. The dataset consisted of 1697 diabetic and 7701 non-diabetic cases, and 13 features, and used synthetic minority oversampling technique (SOMT) to remove the imbalance in dataset. Among ML methods, LR had the highest performance in original data and AdaBoost achieved the highest accuracy and AUC values of 86.2% and 94%, respectively. Based on AdaBoost model, age and years), family history of diabetes, triglycerides and NonHDL were the most important features, respectively. The study indicated that AdaBoost classifier was the best ML-based learning for diabetes prediction, and family history of diabetes.