SDAGS: SMOTE + forest diffusion-based data augmentation and GBT-based stacking ensemble learning for holistic AI-powered diabetes mellitus prediction

Abstract

In emerging nations like Vietnam, it's critical to track and anticipate diabetes in emerging nations like Vietnam, particularly for those with type 1 diabetes. This article proposes SDAPS, an AI-powered diabetes prediction technique. Our method is based on two ideas: (i) using the SFDM method to make the training data better by combining the oversampling of the Forest Diffusion Model with the SMOTE data balancing method; and (ii) making the GSDP model by stacking different boosting machine learning models together. We also suggest an AI-powered blood glucose monitoring and recommendation system based on SDAPS to provide diabetic patients with all-encompassing assistance with blood glucose monitoring, dietary counseling, physical activity, and the proper use of medications. Our thorough experiments using the Pima Indians diabetes dataset and the 5-fold cross-validation method demonstrate that SDAPS outperforms the state-of-the-art methods. Its prediction performance signigicantly achieved a sensitivity of 98.3%, a specificity of 99.49%, an F1-score of 98.74%, an accurary of 98.75%, and precision of 98.00%.