Enhanced CNN–TCN–GTN–SE for Real-Time Heart Rate Classification in Personalized Cardiac Rehabilitation using IoT-driven Smart Wearable Device

Abstract Accurate heart rate monitoring is vital in assessing cardiovascular responses during rehabilitation, particularly for individuals with ischemic heart disease. However, conventional wearable systems often overlook inter-individual variations in heart rate dynamics and personalized cardiac performance, limiting their clinical value. This study presents an enhanced deep learning model that enables real-time classification using an IoT-driven smart wearable device and photoplethysmography (PPG) sensor. The proposed architecture integrates Convolutional Neural Networks (CNN), Temporal Convolutional Networks (TCN), Gated Transformer Networks (GTN), and Squeeze-and-Excitation (SE) blocks to extract both localized and long-range features from heart rate time-series data with [ADAM optimizer]. A total of 1,827 clinically labeled heart rate samples were collected from 38 patients undergoing a structured 480-second rehabilitation protocol. The model was trained and evaluated using 10-fold cross-validation and an independent test set, achieving an accuracy of 98.22% and a specificity of 98.53%. Ablation studies confirmed the contribution of each architectural component. This work introduces a new AI-powered solution, clinically scalable and one of the first models validated on real-world PPG data collected during structured and personalized cardiovascular rehabilitation. This work bridges wearable AI and clinical rehabilitation, offering substantial potential for real-time personalized cardiovascular monitoring and decision support in home-based settings.