Prediction Model for Type 2 Diabetes using Stacked Ensemble Classifiers

Diabetes is the number one of major causes of death globally. Undetected and untreated diabetes causes serious issues and the individuals with diabetes are at high risk for complication. Thus, an early diabetes prediction is necessary to help the individuals preventing dangerous conditions at the early stage. This study proposed a prediction model to offer early prognostication of type 2 diabetes. The proposed model incorporates isolation forest and synthetic minority over sampling-tomek link technique to detect as well as remove the outlier data, and balance the data distribution, respectively. The stacked ensemble classifiers are the used learn and predict type 2 diabetes at an early stage. We used three publicly available datasets to evaluate the performance of proposed model as compared to other models such as multi-layer perceptron, support vector machines, decision tree, and logistic regression. We applied 10-fold cross-validation and obtain four performance metrics such precision, recall, f-measure, and accuracy. The experimental results show that the proposed model outperformed other models, achieving accuracy up to 93.18%, 98.87%, and 96.09% for dataset I, II, and III, respectively. It is expected that the early diabetes prediction could help the individuals on taking precautions once type 2 diabetes is detected.