Enhancing Automatic Detection of Frustration Induced During HCI with Moment-based Biosignal Features

摘要

Enhancing HCI systems with the capability to detect user's frustration and respond appropriately is a significant challenge. In this line, biosignal features based on the theory of orthogonal Krawtchouk and Legendre moments are assessed in the present work over their ability to enhance accuracy in automatic detection of frustration, which is induced through HCI, during video-game playing. Experimental evaluation, conducted over a multi-subject dataset over frustration detection showed that conventional features, typically extracted from Galvanic Skin Response and Electrocardiogram in the past, achieved correct classification rate (CCR) of 83.59%. Fusing these conventional features with moment-based ones extracted from the same modalities resulted to significantly higher accuracy, at the level of 93%. Furthermore, moment-based features lead also to over 10% increase in CCR when the aim was to identify both bored and frustrated cases,within a 3-classs affect detection problem.