An Empirical Overview of Stress Detection Based on Photoplethysmography Signal Using Various Deep Learning Techniques

The rapid speed of modern life and the shift towards cognitive work have contributed to a rising prevalence of mental stress across all professions. Many sedentary occupations, particularly in fields like Information Technology (IT), causes prolonged periods of intense mental effort that can significantly elevate stress levels. Extended exposure to such mental stress is linked to an increased risk of serious conditions, such as cardiovascular diseases and mental health disorders. Consequently, there is a critical need for non-intrusive tools capable of detecting mental stress. Early detection is essential for preventing stress-related health concerns and continuous observation of stress levels. Numerous methods are currently available for stress detection, but they face limitations like inaccuracies due to individual differences and environmental factors as well as difficulties with real-time monitoring. In this discussion, several techniques are explored that offer better insights with stress detection, providing valuable perspectives for future researchers. This survey discusses various deep learning techniques, highlighting their advantages and limitations. It also suggests potential improvements to tackle these limitations in the future. Initially, this survey categorizes Deep Learning techniques into five techniques: Long ShortTerm Memory (LSTM), Convolutional Neural Network (CNN), Transfer Learning (TL), Attention-Based Mechanisms, and Transformer-based approaches are discussed in depth. Subsequently, it identifies research gaps in traditional Deep Learning approaches to guide future innovative model development. The study reviews 25 research papers focusing on various methods for stress detection using DL. Finally, this survey evaluates performance metrics, datasets, techniques used for research, and model achievements. CNNs emerge as the predominant approach, and Python is frequently used for stress detection in DL. Additionally, accuracy is highlighted as a key analytical metric.