Regression-based approach to anxiety estimation of spider phobics during behavioural avoidance tasks

Florian Grensing, Vanessa Schmücker, Anne Sophie Hildebrand, Tim Klucken, Maria Maleshkova · 2025

Read Regression-based approach to anxiety estimation of spider phobics during behavioural avoidance tasks by Florian Grensing, Vanessa Schmücker, Anne Sophie Hildebrand, Tim Klucken, Maria Maleshkova (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-13795.txt.

Introduction

Florian Grensing and co-authors estimate spider-phobic anxiety from wrist sensors during exposure tasks, in work from 2025. Questionnaires and avoidance tests capture only snapshots of fear, while therapy could use a continuous readout.

Twenty-five participants faced four behavioral avoidance tasks in one session, reporting their anxiety at intervals. Heart rate, heart variability, skin conductance, and skin temperature fed regression models. Models using only body signals did respectably. Adding computed signal features helped, and adding context about the current task helped most, cutting prediction error sharply. The authors say wearable sensing can track anxiety closely enough to assist therapy planning and personalize treatment.

Catalog record

CREATORFlorian Grensing, Vanessa Schmücker, Anne Sophie Hildebrand, Tim Klucken, Maria Maleshkova
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-13795

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Regression-based approach to anxiety estimation of spider phobics during behavioural avoidance tasks Florian Grensing, Vanessa Schmücker, Anne Sophie Hildebrand, Tim Klucken, Maria Maleshkova Abstract Phobias significantly impact the quality of life of affected persons. Two methods of assessing anxiety responses are questionnaires and behavioural avoidance tests (BAT). While these can be used in a clinical environment they only record momentary insights into anxiety measures. In this study, we estimate the intensity of anxiety during these BATs, using physiological data collected from unobtrusive, wrist-worn sensors. Twenty-five participants performed four different BATs in a single session, while periodically being asked how anxious they currently are. 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) using only physiological signals, (2) adding computed features (e.g., min, max, range, variability), and (3) computed features combined with contextual task information. Adding contextual information increased the effectiveness of the model, leading to a root mean squared error (RMSE) of 0.197 and a mean absolute error (MAE) of 0.041. Overall, this study shows, that data obtained from wearables can continuously provide meaningful estimations of anxiety, which can assist in therapy planning and enable more personalised treatment. K eywords First keyword \cdot Second keyword \cdot More 1 Introduction Phobias are intense and irrational fears of specific objects, situations, or activities that significantly impact an individual’s daily life [ 17 ] . Common phobias include arachnophobia (fear of spiders), acrophobia (fear of heights), and claustrophobia (fear of confined spaces) [ 17 ] . A central component in the treatment of phobias is exposure therapy [ 11 ] , in which individuals are gradually exposed to the feared stimulus in order to reduce their anxiety over time. A commonly used tool to assess phobia severity in clinical and research settings is the Behavioural Avoidance Test (BAT) [ 2 ] . In this task, individuals are asked to approach a fear-inducing stimulus, such as a spider, as closely as possible, but are allowed to stop whenever they wish. The point at which they stop is recorded as a behavioural measure of anxiety intensity. Variants of the BAT include walking toward the stimulus [ 16 ] , pulling it towards oneself [ 21 ] , or interacting with it in virtual reality (VR) environments [ 16 ] , which allow for more controllable and replicable exposure scenarios. Although the BAT is not typically part of exposure therapy itself, it is frequently used to evaluate treatment outcomes before and after therapeutic interventions [ 3 ] . While the BAT provides valuable insights into anxiety behaviour, it does not offer insight into the fluctuations of anxiety during the task or throughout treatment. Recent advancements in physiological monitoring offer new methods for objectively assessing fear and anxiety. Measures such as heart rate and skin conductance have been shown to correlate with emotional arousal, and could provide a more continuous, real-time estimate of anxiety during exposure sessions [ 24 ] [ 22 ] . In this study, we investigate how physiological signals can be used to estimate moment-to-moment anxiety intensity of subjects with a fear of spiders. Specifically, we aim to model self-reported anxiety during BATs using data such as heart rate and skin conductance. Our approach offers a step toward more objective, data-driven methods for monitoring patient responses during exposure therapy. To this end, we collect and analyse physiological and self-report data during BATs and develop machine learning models that estimate continuous anxiety intensity. The contributions of this study are (1) the collection of data from 25 participants during four Behavioural Avoidance Tests, (2) estimation of continuous anxiety intensity using wrist-worn wearables, and (3) integration of contextual information into the machine learning models. This paper is structured as follows: First, we present some background on related works. Afterwards, we describe the study setup, as well as the data collection and preprocessing. Next, the results will be presented and discussed. Finally, the results are summarised and an outlook for future research is presented. 2 Research Background Fear responses are mediated by the autonomic nervous system (ANS), particularly the sympathetic division, which triggers the “fight-or-flight” response [ 12 ] . This then triggers various physiological responses, which can be used for emotion detection. Although works, such as that of Ekman et al. [ 6 ] suggest using camera-based facial recognition, this is often difficult in combination with technologies like virtual reality, which are becoming increasingly popular in the field of therapy. The same is true for electroencephalography (EEG), which, while informative, is intrusive and can be difficult to integrate into VR-based applications. Therefore, this paper specifically focuses on physiological signals that are measurable by using wrist-worn wearables, such as heart rate (HR), electrodermal activity (EDA), also known as galvanic skin response (GSR) and body temperature. These signals have successfully been used for emotion detection in the past. EDA has been found suitable for emotion classification among both the valence and arousal dimensions of emotion recognition [ 18 ] . A study by Polo et al. has used EDA to distinguish between sadness, happiness, fear and relaxation [ 18 ] . Ménard et al. have used a combination of EDA and heart rate, to classify emotions [ 15 ] . Participants were shown video clips to elicit different emotions, which were then classified with an accuracy of 85% for EDA and 89% for heart rate respectively. Other signals, such as heart rate variability, breathing rate, blood pressure have also been used [ 5 ] . One study that used physiological signals to estimate anxiety during virtual reality therapy was performed by Freeman et al. [ 4 ] and used VR for treating fear of heights. During the experiment, EEG, GSR and respiration data were collected. Sixty-one participants performed a relaxation–stress–relaxation type tasks multiple times over four weeks. Using the collected physiological data, the authors achieved an 88% accuracy for a binary classification using a support vector machine (SVM). In 2021, Šalkevicius et al. presented their work, in which they collected BVP, GSR and skin temperature data during a VRET and used that data to predict the participants anxiety levels. They split the anxiety into 4 classes (‘low’, ‘mild’, ‘moderate’, and ‘high’) and performed classification. Across 30 participants, they have attained an accuracy of 80.1% [ 19 ] . While these studies have demonstrated that physiological signals can be used to classify emotional states, relatively few have focused on predicting continuous, moment-to-moment anxiety intensity using wearable data. Šalkevicius et al. have incorporated Subjective Units of Distress Scale (SUDS) ratings in the context of virtual reality exposure therapy (VRET), they have discretized these scores into categorical anxiety levels (e.g., low, mild, moderate, high). In contrast, the present study uses regression to predict continuous SUDS scores, allowing for a more fine-grained analysis of anxiety fluctuations over time. 3 Methods In the following section, we describe the experiment design, including participant recruitment, the four Behavioural Avoidance Tests, the preprocessing of the physiological data and the machine learning models used for the anxiety prediction. Participants: The study involved 25 participants (3 male, 22 female) with an average age of 24.96 years (SD = 8.75). All participants met

Read the complete text