Time Series Modeling of Electrodermal Signals to Understand Cognition During Gaming

Electrodermal activity (EDA) is a measure of change in the electrical activity due to skin conductance of the skin. It has found its application in determining psychological changes such as changes in emotions or cognitive load. Despite EDA signals being less complex than electromyography and electroencephalography signals, the application of the EDA in real-time applications is limited. This study explores the change of EDA signals with time to determine cognitive load using a pattern recognition approach. In this study, the electrodermal signals are acquired from different participants performing a video gaming task. The EDA signals acquired are subjected to the pattern recognition process including computation of features, feature optimization using ReliefF algorithm and regression modelling. 11-time domain features were subjected to feature ranking using ReliefF. The algorithms used for regression modelling include Exponential Gaussian, Support Vector Machines and Random Forest. The features ranked highest included Waveform Length (WL), Maximum Value, 3rd, 4th and 5th temporal moments. The mean R-square values for the regression model obtained in this study were 0.81 and 0.8 using exponential Gaussian and Random Forest Models respectively. The subject-specific pattern recognition analysis showed an R-square of 0.99 in certain subjects.