EDGE OF THINGS BASED DIABETES PREDICTION USING DEEP LEARNING

Abstract

Diabetes is a chronic disorder that is very prevalent in the world and the severe complications that can be caused through this disorder include heart disease, kidney failure, loss of vision, and damage of nerves. Since the condition cannot be fully reversed, the early recognition of the condition is critical in reducing the chances of its effects in the long run, as well as enhancing the day-to-day quality of life of the patients. The health care that is present now-a-days is still characterized by manual diagnoses and hospital based tests which cause delays and increases the costs. It is now possible to constantly monitor health and predict the risk of disease in real-time due to the emergence of the Internet of Things (IoT) and Edge Computing.

The current paper introduces an Edge of Things (EoT) that predicts diabetes with the help of ensemble and deep learning. Its framework has consumed the data-preprocessing phases of Z-score normalization, SMOTE, ADASYN, SMOTE-ENN, and PROWRAS to at least create a balanced set of data and uses Principal Component Analysis (PCA) in feature selection. It develops and tests three hybrid deep-learning models, including LeNetEncoder, TCNEncoder and GRUEncoder. The machine-learning stacking model is also constructed with Base learners that include Decision Tree, Random Forest, and Extra Trees, and the meta -learner, LightGBM. The effectiveness of the method will be checked using a five-fold cross-validation strategy. Findings demonstrate that the suggested models are much more accurate, more precise, more recall-oriented, as well as more F1-score oriented, when compared to the traditional machine-learning classifiers.