Affiliations: Information and Human Centred Computing Research Group, Research School of Computer Science, Australian National University, Canberra, ACT, Australia
Correspondence:
[*]
Corresponding author: Nandita Sharma, Information and Human Centred Computing Group, Research School of Computer Science, Building 108, Australian National University, Canberra, ACT 0200, Australia. Tel.: +1 612 6125 9664; E-mail: nandita.sharma@anu.edu.au
Abstract: Stress is a major problem in our society today and poses major concerns for the future. It is important to gain an objective understanding of how average individuals respond to events they observe in typical environments they encounter. We developed a computational model of stress based on objective human responses collected from human observers of environments. In the process, we investigated whether a computational model can be developed to recognize observer stress in abstract virtual environments (text), virtual environments (films) and real environments (real-life settings) using physiological and physical response sensor signals. Our work proposes an architecture for a computational observer stress model. The architecture was used it to implement models for the different types of environments. Sensors appropriate to the different types of environment were investigated where the aims were to achieve unobtrusive methods for stress response signal collection, reduce encumbrance and hence, enhance methods to capture natural observer behaviors and produce stress models that recognized stress more robustly. We discuss the motivations for each investigation and detail the experiments we conducted to collect stress data sets for observers of the different types of environments. We describe individual-independent artificial neural network and support vector machine based model classifiers that were developed to recognize stress patterns from observer response signals. The classifiers were extended to include a genetic algorithm which was used to select features that were better for stress recognition and reduce the use of redundant features. The outcomes of this research provide a possible future extension on managing stress objectively.
