AI-Driven diabetes prediction and early diagnosis system using wearable sensors and real-time health monitoring

Abstract

Diabetes Mellitus (DM) is a metabolic disorder characterized by high blood sugar levels, which have been persistent for a long period of time, resulting from the way the pancreas functions, either by the release of insufficient amounts of insulin or the proper use of insulin by the body, or a combination of both. In recent times, the global burden of diabetes has increased significantly, with about 463 million people worldwide living with diabetes by the end of 2019, a figure expected to rise to 700 million by the end of 2045. In order to provide a proactive system of health care, the system has been designed to continuously monitor physiological data, use AI for the analysis of health trends, and send real-time alerts for the early detection of diabetes. In the development of the system, a systematic approach was adopted, which integrated wearable sensors, cloud computing, and machine learning techniques. Five-fold cross-validation was carried out, and the models, including Random Forest, Gradient Boosting, Logistic Regression, and SVM, were trained. Accuracy, F1 score, AUC, and confidence were the evaluation metrics for the models, which showed high predictive capabilities, with the highest accuracy of 94.67% obtained by the tree-based models, which were further optimized. Ensemble methods, including a soft voting classifier, were also developed, which showed reliable and robust performance for the early detection of diabetes, indicating that wearable sensors integrated with AI have the potential to revolutionize the health care of diabetes patients.