A Novel Deep Learning-Based Classification Model For ECG-Derived Stress Recognition

Abstract

Stress recognition using electrocardiogram (ECG) signals is a promising approach for real-time health monitoring, offering a non-invasive, efficient, and cost-effective solution for detecting stress in individuals. Since stress can have detrimental effects on both physical and mental health, early detection is critical for preventive healthcare. Hence, the proposed model intends to improve a robust model for stress recognition by analyzing ECG signals collected under different stress-inducing conditions. This research utilizes a Kaggle dataset, which includes features from the Stress and Workload Evaluation of Lifelong Learning (SWELL) and Wearable Stress and Affect Detection (WESAD) datasets. To enhance signal quality and ensure accurate analysis, preprocessing techniques using StandardScaler are applied to standardize feature distributions. The significant features were extracted using a novel framework called Weighted Adaptive Hybrid Independent Discriminant Linear Extraction (WAHIDLE), integrating Weighted Independent Component Analysis (WICA) and Adaptive Linear Discriminant Analysis (ALDA). For classification, a novel Ensemble Randomized Support Vector Convolutional Neuro Net (ERSV-ConvNet) is employed to categorize signals into stress and non-stress categories. A range of Python-based tools and libraries, combined with comprehensive comparative analyses, are utilized to demonstrate the effectiveness and robustness of the proposed method. The results indicate that the approach outperforms existing stress recognition models, delivering improved accuracy (99.98%), and efficiency. The findings contribute to the development of personalized health applications and wearable devices capable of real-time stress detection, empowering individuals to manage stress more effectively and mitigate its negative health impacts.