The Role of Features Types and Personalized Assessment in Detecting Affective State Using Dry Electrode EEG
| Autor: | Emmanuel Rios Velazquez, Paruthi Pradhapan, Yelena Tonoyan, Vojkan Mihajlovic, Jolanda Witteveen |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2020 |
| Předmět: |
Computer science
Entropy media_common.quotation_subject Emotions 02 engineering and technology Electroencephalography Affect (psychology) lcsh:Chemical technology Biochemistry Article Analytical Chemistry Task (project management) Arousal 03 medical and health sciences 0302 clinical medicine arousal Perception 0202 electrical engineering electronic engineering information engineering medicine Humans lcsh:TP1-1185 EEG Electrical and Electronic Engineering Valence (psychology) valence dry electrodes Electrodes Instrumentation media_common medicine.diagnostic_test human affect wearable EEG Atomic and Molecular Physics and Optics machine learning Binary classification 020201 artificial intelligence & image processing 030217 neurology & neurosurgery Cognitive psychology |
| Zdroj: | Sensors, Vol 20, Iss 6810, p 6810 (2020) Sensors (Basel, Switzerland) Sensors Volume 20 Issue 23 |
| ISSN: | 1424-8220 |
| Popis: | Assessing the human affective state using electroencephalography (EEG) have shown good potential but failed to demonstrate reliable performance in real-life applications. Especially if one applies a setup that might impact affective processing and relies on generalized models of affect. Additionally, using subjective assessment of ones affect as ground truth has often been disputed. To shed the light on the former challenge we explored the use of a convenient EEG system with 20 participants to capture their reaction to affective movie clips in a naturalistic setting. Employing state-of-the-art machine learning approach demonstrated that the highest performance is reached when combining linear features, namely (a) symmetry features and single-channel features, with nonlinear ones derived by a multiscale entropy approach. Nevertheless, the best performance, reflected in the highest F1-score achieved in a binary classification task for valence was 0.71 and for arousal 0.62. The performance was 10&ndash 20% better compared to using ratings provided by 13 independent raters. We argue that affective self-assessment might be underrated and it is crucial to account for personal differences in both perception and physiological response to affective cues. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|