A Systematic Review on Physiology-based Anxiety Detection using Machine Learning.

Abstract

Anxiety disorder poses a significant challenge to mental health. Diagnosing anxiety is complicated due to its various symptoms and factors, often resulting in extended periods of untreated patient suffering. As a result, patients often endure prolonged periods without treatment. This scenario has prompted researchers to step into the domain of non-invasive physiological signals, including electroencephalography, electrocardiogram, electromyography, electrodermal activity, and respiration. By integrating machine learning into the physiological signals, clinicians can identify distinct anxiety patterns and effectively differentiate between individuals with the disorder and those in good health. This paper presents a systematic literature review of physiological sensors and machine learning methods to diagnose and predict anxiety disorder. It also presents an overview of wearable devices employed in previous studies. A key contribution of this review is the exploration of the relationship between physiological features and anxiety disorders through machine learning models. The paper discusses methodologies, open datasets, and identifies research gaps and challenges related to the machine learning-based analysis of physiological signals for anxiety detection. Furthermore, a novel multimodal approach for anxiety classification is proposed, utilizing a combination of physiological signals. This review aims to provide a comprehensive understanding of the current trends, architectures, and techniques employed in the field of anxiety detection.