CNN-LSTM Hybrid Framework for Deep Learning-Based Stress-Aware Arrhythmia Classification on ECG Data

A major difficulty in proactive cardiac care is addressed in this research by proposing a CNN-LSTM hybrid framework for identifying and categorizing stress-induced arrhythmias from ECG data. The approach uses three-fold time series cross-validation, dual-path feature extraction, robust modelling with data augmentation, and extensive preprocessing (Butterworth filtering, wavelet denoising, and Z-score normalization) to use datasets such as the MIT-BIH Arrhythmia and Noise Stress Test. The model performs exceptionally well, achieving 95.48% accuracy and 95.13% F1-score in arrhythmia classification and 99.21% accuracy, precision, recall, and F1-score in stress detection. It performs better than current models because it successfully integrates variability associated with stress into ECG analysis. The model is ideal for real-time monitoring on edge devices and wearables since it is practical, has low-latency inference (45 ms per segment), and uses little memory (< 60 MB). This system allows constant cardiac monitoring in practical healthcare applications, facilitates early identification, and enhances patient outcomes.