Machine Learning for Anxiety Detection Using Biosignals: A Review.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Psychiatry

Background:

• Anxiety disorder (AD) diagnosis is challenging due to diverse symptoms and confounding factors, leading to delayed treatment.
• Non-invasive biosignals including electroencephalography (EEG), electrocardiogram (ECG), electrodermal response (EDA), and respiration (RSP) are increasingly explored for anxiety detection.
• Machine learning (ML) offers potential for analyzing complex biosignal patterns to aid in differentiating anxiety patients from healthy individuals.

Purpose of the Study:

• To review and summarize studies from 2012-2022 on ML algorithms applied to biosignals for anxiety detection.
• To analyze the strengths and weaknesses of current ML-based anxiety detection methods.
• To provide insights for future advancements in non-invasive anxiety diagnosis.

Main Methods:

• Systematic literature review of studies published between 2012 and 2022.
• Analysis of various biosignals (EEG, ECG, EDA, RSP, heart rate) and ML algorithms (Random Forest, Support Vector Machines, Neural Networks).
• Evaluation of measurement accuracies, sample sizes, and combinations of biosignals.

Main Results:

• Promising measurement accuracies ranging from 55% to 98% were reported across studies with 10-102 participants.
• While EEG-only studies showed good performance, combined EDA, RSP, and heart rate yielded the most accurate results.
• Random Forest and Support Vector Machines require feature selection, whereas Neural Networks offer good accuracy without it.

Conclusions:

• ML analysis of biosignals presents a viable approach for anxiety disorder detection.
• Effective combinations of biosignals and appropriate ML models significantly enhance diagnostic accuracy.
• Further research can leverage these findings to develop more convenient and accurate anxiety detection tools.