Detection of universal cross-cultural depression indicators from the physiological signals of observers

J. F. Plested, T. D. Gedeon, X. Y. Zhu, A. Dhall, R. Geocke

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearch

2 Citations (Scopus)

Abstract

We conducted a pilot study experimenting with neural network techniques to use the physiological signals of untrained observers to classify the depression levels of variously depressed people in videos speaking a language the observers did not understand. As the dataset was highly imbalanced, noisy and thus extremely sensitive to relative class sizes, we developed a technique for dynamically oversampling the smaller classes both prior to and during training to approximately align training prediction rates for each class with knowledge of the prevalence of different levels of depression. In predicting the depression levels to a final accuracy of 57.9% over four classes and 78.9% over three classes we demonstrate the likelihood that universal cross-cultural indicators of depression exist. In addition, that some people's automatic physiological responses to these indicators are strong enough that they can be used to predict depression categories of people to a significant degree of accuracy even when the observer does not understand the language the person is speaking. The final accuracy rate is significantly better than the diagnosis rates of doctors speaking to patients in their own language. The results show the potential these techniques have to improve diagnosis of depression, especially in areas with limited access to mental health professionals. This innovative approach demonstrates the importance of further experimentation in this area and research into universal cross-cultural depression indicators.

Original languageEnglish
Title of host publication2017 Seventh International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW 2017)
EditorsHayley Hung, Emily Mower Provost, Mohammad Soleymani
Place of PublicationPiscataway NJ USA
PublisherIEEE, Institute of Electrical and Electronics Engineers
Pages185-192
Number of pages8
ISBN (Electronic)9781538606803
ISBN (Print)9781538606810
DOIs
Publication statusPublished - 2017
Externally publishedYes
EventInternational Conference on Affective Computing and Intelligent Interaction Workshops and Demos 2017 - San Antonio, United States of America
Duration: 23 Oct 201726 Oct 2017
Conference number: 7th
http://www.lab-msp.com/ACII2017org/workshops

Conference

ConferenceInternational Conference on Affective Computing and Intelligent Interaction Workshops and Demos 2017
Abbreviated titleACIIW 2017
CountryUnited States of America
CitySan Antonio
Period23/10/1726/10/17
Internet address

Cite this

Plested, J. F., Gedeon, T. D., Zhu, X. Y., Dhall, A., & Geocke, R. (2017). Detection of universal cross-cultural depression indicators from the physiological signals of observers. In H. Hung, E. Mower Provost, & M. Soleymani (Eds.), 2017 Seventh International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW 2017) (pp. 185-192). IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ACIIW.2017.8272612