ReBAT: Regression-Based Anxiety Recognition during Be-havioural Avoidance Tasks

¹ Professorship of Data Engineering, Helmut Schmidt University, Hamburg, Germany
² Department of Clinical Psychology, University of Siegen, Siegen, Germany
³ Clinic for Psychiatry and Psychotherapy, University Hospital OWL, Bielefeld University, Bielefeld, Germany

^* e-mail: Grensinf@hsu-hh.de
^** e-mail: Vanessa.Schmuecker@uni-siegen.de
^*** e-mail: Anne.Hildebrand@evkb.de
^**** e-mail: Tim.Klucken@psychologie.uni-siegen.de
^† e-mail: Maleshkm@hsu-hh.de

Abstract

Behavioural Avoidance Tests (BATs) are commonly used to assess anxiety responses in individuals with specific phobias, but they typically rely on observable behaviour and self-report, offering only momentary insights into anxiety levels. This study investigates the feasibility of continuously estimating anxiety intensity during BATs using physiological data collected from a wrist-worn sensor. Twenty-five participants with spider phobia completed four BATs, both in vivo and virtual reality, in a single session. Using heart rate, heart rate variability, electrodermal activity, and skin temperature, we trained regression models to predict anxiety ratings from three types of input data: (1) only physiological signals, (2) computed features (e.g., min, max, range, variability), and (3) computed features combined with contextual task information. Predictive performance improved with added feature complexity, with the best model achieving a root mean squared error (RMSE) of 0.197 and a mean absolute error (MAE) of 0.041. The results clearly show that wearable sensors can provide meaningful, continuous estimations of anxiety during BATs, which can assist therapists in therapy planning to create more personalised treatment.