Design Engineering
Showcase 2020

Social anxiety is defined as a fear of social situations, where the individual might be exposed to possible scrutiny, it typically causes physiological changes in the affected person. Unfortunately, although there is a high prevalence of social anxiety in young people at a subclinical level, many do not receive treatment. This is due to lack of treatment available and the lack of recognition by healthcare professionals. Some may not even seek treatment due to the fear of being negatively evaluated.

If left untreated, one can develop social anxiety disorder, one of the most common anxiety disorders; as well as comorbidities such as depression which cause even further life impairments and require more costly treatment.

Due to the rising prevalence of social anxiety, current approaches are inadequate. Detection of social anxiety using wearables may provide a novel way of recognising it, which reveals new opportunities for monitoring and treatment. For example, this method of detection could reduce economic impact and strain on mental health services, by potentially facilitating treatment in a digital self-help manner – offering treatment at earlier stages when it is less costly and extensive. It could also be argued that this approach to treatment could be more comfortable for sufferers due to their fear of being negatively evaluated.

Therefore, this project investigates whether social anxiety in young people can be detected using physiological data collected from a wearable, by combining knowledge from machine learning and psychology fields.

Physiological data was recorded while thirteen young people with social anxiety participated in impromptu speech tasks. Following a supervised machine learning approach, various classification algorithms were used to develop models for three different contexts, investigating the detection of social anxiety and its nature. The models were then evaluated using 10-fold cross validation. Focus groups were also conducted with socially anxious young people to further evaluate the impact of the study and research in this area.

The findings were promising. All classification models in all machine learning experiments (detection of presence of social anxiety, its nature and severity) yielded above 90% accuracy. The data analysis also indicated that physiological changes correlated with severity. Sweat response was also identified as the most effective singular modality when differentiating between baseline and social anxiety states, whereas skin temperature was the most effective modality when classifying severity.

Additionally, the focus groups further emphasised the positive impact of the study and the need for further research in this area. This could transform the current approaches to diagnosing, treating and monitoring social anxiety which could have great social and economic impact.

If you would like to know more about the project feel free to get in contact.