Machine Learning-Based Detection of High Trait Anxiety Using Frontal Asymmetry Characteristics in Resting-State EEG Recordings

Due to the rising prevalence and severe influence on society’s physiological and mental health, anxiety has emerged globally as one of the most severe disorders. While delayed detection and treatment can lead to fatal consequences for an individual’s health, preventive interventions are required to identify and treat the disease prematurely. Detecting trait anxiety could significantly reduce the prevalence of anxiety disorders. This study proposes a machine learning model that uses unfolded EEG spectra and frontal asymmetric brain activity characteristics to diagnose anxiety tendencies. With our approach, we identified the most predictive electrodes and the corresponding frequency subbands to distinguish between low and high levels of trait anxiety in individuals (F8/F7 9.110 Hz, 16.117 Hz; FC6/FC5 5.16 Hz; AF4/AF3 18.119 Hz; F2/F1 33.134 Hz). By achieving a balanced accuracy of 81.25 percent, our novel detection approach represents a benchmark in objectively detecting different stages of trait anxiety.