User Stress Modeling through Galvanic Skin Response

The advent of digital era has brought great advances in the quality and accuracy of Bio medical sensors and other physiological devices. Similarly, digital games have also witnessed massive improvements in their scale, mechanics, graphics, and reach, which has led to a fierce debate on their human and societal impact, especially in terms of identifying the correlation, if any, between the gamer and violent transgressors. From a pure technological perspective, it is thus imperative that advances in sensory technologies and machine learning are then utilized to build a model for identifying the stress experienced by the gamer, during any game session. Galvanic Skin Response(GSR), can act as a good indicator of this experienced stress, by measuring the change in skin conductance and skin resistance of the user. However, GSR data, in its raw form, is very much user dependent, often biased, and is difficult to analyze, as it gives a long term measure of the user behavior changes, based on skin precipitation. In this research work, we have collected user's perceived notion of stress along with sensory data from a GSR device, which was then analyzed using various machine learning models, before creating a majority voting based ensemble model for stress modeling. Showing comparable values of accuracy(63.39%) and precision(51.22%), our model was able to substantially increase the class recall rate for identifying stress (27.08%), from the individual approaches (0-8.95%).