Exploring Skin Potential Signals in Electrodermal Activity: Identifying Key Features for Attention State Differentiation

Abstract

Assessing changes in human attention states via noninvasive physiological signals poses a significant challenge. Research has often associated the migration of mitochondrial superoxide radicals with electrical skin signals and physiological state. This study proposes a novel method to identify meaningful features for discerning variations in attention states by examining skin potential (SP) signals from specific features. This research project began with gathering SP signals. Next, Wavelet Packet Transform (WPT) and other approaches are applied to conduct an energy analysis across various frequency bands, which allows for the extraction of time- and frequency-domain features from the SP signals, and the potential of these features to differentiate human attention states is then examined via regression-based classifiers. Feature selection refinement is accomplished through statistical tests and Linear Support Vector Machines (SVM) with Recursive Feature Elimination (RFE). The focus is on the discriminative power of a selected set of primary features to distinguish human attention states. The designed experiment revealed significant variations in SP signal features when the subjects experienced shifts in their attention states. These features encompass measures such as first- and second-order derivative sequences, wavelet energy, wavelet coefficient, and power spectral density in different frequency bands. The core significance of this research lies in its focus on the feature selection of the SP signal, which yields a set of highly impactful features contributing to the distinction of attention states. This study underscores the potential of classifier models to effectively distinguish attention states, particularly through the examination of critical features, such as the wavelet energy of SP signals within certain frequency bands. These features may be relevant to several psychological mechanisms that reinforce the relationship between physiological signals and cognitive state. The insights derived from this investigation deepen the comprehension of human attention states and set the groundwork for more granular future explorations of SP signals.

Keywords

• Electrodermal activity
• feature selection
• machine learning
• SP signal
• attention
• wavelet