EDGE OF THINGS BASED DIABETES PREDICTION USING MACHINE LEARNING

Abstract

Diabetes mellitus is a fast-increasing health issue in the world which needs to be diagnosed and managed in time to minimize complications and health expenses. Recent developments in machine learning have shown good prospects towards enhancing diabetes prediction, but, the majority of those existing solutions are based on centralized cloud models which are characterized by a great latency, privacy reasons, and inability to use them in resource-limited settings. This paper attempts to deal with these issues by suggesting an Edge of Things (EoT)-based diabetes predictive model based on the integration of hybrid deep learning models with ensemble machine learning algorithms to achieve decentralized and real-time prediction of diseases. The suggested structure will include extensive data preparation, features normalization, and the class imbalance to optimize predictive accuracy. Hybrid deep learning models are used to learn the complex nonlinear association among demographic, clinical and behavioral variables and ensemble learning methods exploit the synergistic advantages of multiple classifiers to enhance their robustness and generalization. The system will greatly decrease the response latency and bandwidth consumption, and will avoid relying on constant internet connection, which will also increase data privacy and security by deploying trained models on the edge. The experimental assessment of the publicly available diabetes data base shows that the hybrid-ensemble framework proposed is more accurate, sensitive, and stable as compared to the conventional one-model frameworks. The findings indicate that imbalance-conscious learning and edge-based intelligence is effective in healthcare analytics. On the whole, this research provides a solution to detecting diabetes at an early stage saving privacy and being scalable and efficient to support the proactive and individualized healthcare delivery in resource-restricted conditions.