Predicting Short-Term Glucose Trends in Real Time Using Multimodal IoT Data and Neural Networks

Abstract—The high prevalence of diabetes worldwide has made it necessary to have a continuous, non-invasive, and accurate way to measure blood glucose levels in order to prevent complications and improve quality of life. The current study proposes a sophisti- cated real-time blood glucose prediction model based on Internet of Things (IoT) and artificial intelligence (AI) technologies. Real- time physiological data, such as heart rate, skin temperature, and activity levels, are measured by CGM and wearable IoT devices and sent to an edge-based AI model for analysis. Machine learning and deep learning algorithms like LSTM and GRU are used to learn temporal relations and produce highly accurate short-term glucose variation forecasts. For the system to operate smoothly in the context of both clinical and personal healthcare, low latency, energy efficiency, and scalability are key design considerations. The experimental results show a mean absolute error (MAE) of 7.6 mg/dL, a prediction error of greater than 92 ratio, and a shorter latency of less than 1.5 seconds per prediction. The article outlines the potential for combining AI and IoT to help with proactive diabetes management, which would enable it to react quickly to alerts, provide a customized treatment plan for each patient, and apply therapeutic options that lower health risks. Index Terms—Real-time prediction, blood glucose, AI, IoT, wearable sensors, deep learning, diabetes management, continu- ous monitoring.