OPTIMIZING IOMT BASED SMART HEALTHCARE WITH DEEP EHR-TANFLOW: A HYBRID ATTENTION DRIVEN MODEL

Abstract

The increasing demand for efficient and accurate disease diagnosis in Internet of Medical Things (IoMT)-based smart healthcare systems necessitates advanced predictive models that capture complex temporal dependencies in Electronic Health Records (EHR) data. However, existing approaches often struggle with long-term dependency modelling, feature redundancy, and computational inefficiency, which limits their applicability in real-world clinical settings. To address these challenges, this research introduces the Deep Learning EHR-Based Temporal Attention Flow (Deep EHR-TANFlow) framework. This novel architecture processes structured EHR data without requiring direct physical intervention. The framework begins with a robust preprocessing pipeline that incorporates KNN imputation for missing values, Min–Max scaling for normalization, and a sliding-window strategy to generate refined temporal segments. This ensures high-quality, structured input suitable for effective temporal predictive modelling. At its core, the framework employs the CNN–BiGRU Temporal Attention Network (CG-TAN), developed specifically for IoMT-based healthcare monitoring. The model integrates Convolutional Neural Networks (CNN) for local feature extraction, Bidirectional Gated Recurrent Units (Bi-GRU) for modelling forward and backward temporal dynamics, and a Self-Attention (SA) mechanism to enhance global contextual representation. This hybrid architecture effectively captures both local and global dependencies within patient data, significantly improving disease prediction and risk assessment. Experimental results demonstrate that the proposed system achieves superior diagnostic performance, attaining 98% accuracy, 98.12% precision, and 98.08% F1-score. These findings highlight the resilience, efficiency, and low computational complexity of the Deep EHR-TANFlow framework, making it a dependable solution for IoMT-based continuous health monitoring and automated risk assessment while ensuring reliable information extraction and enhanced patient safety.