Quantitative analysis of affective states based on psycho-physiological research using biosignal processing technique

Abstract

The objective of this thesis is to investigate human emotion and cognitive states based on the psycho-physiological understanding using various biosignals. Three experiments were performed by participant with independent experimental environment and protocols. These results and expected effects were summarized as follows.(1) The study of human emotion was shown the difference for the responses of event related potential (ERP) from participants (n=19, 30 age) through two kinds of protocols (picture: number set with emotional faces: cognitive responses depending on emotion distractor, video clips: sadness/amusement emotion stimuli: cognitive responses with emotion mood as interference).The first experiment has statistically approached the ERPs (n75, p100, n200, and p300) by correspondence with emotional faces from EEG signal. The ERPs were relatively more influenced by negative emotional faces such as annoy, disgust, and sadness. The second experiment demonstrate that negative emotion (sadness) could has an affected on the cognitive abilities (ERP, behavioral response) as interference, and these response has a trend depending on both gender and disposition. (2) The developed method, genetic algorithm based optimized extreme learning machine, demonstrate that this algorithm has resulted in the high accuracy (87.9%), the selection of dominated EEG features, and optimized algorithm structure for classifying the mental states.(3) Sleep study was investigated compromising factors between the accuracy for estimating the sleep stage and the convenience for measuring signal. EEG is recommended for applications that require high accuracy (85%), whereas for applications that prioritize ease of measurement rather than accuracy, use of the autonomic nervous system related signal combination is recommended (80%).From the three independent experiments, I expect these results have a positive spreading effect. (1) These results have shown the possibility of quantitative analysis for estimating human affective states from biological features and advanced algorithms. These results and methods are available in psychology, medicine, ergonomics, and forensic science. (2) The developed algorithm with attention (cognition) could be applicable to education and medicine (e.g. ADHD). Finally, (3) we could indirectly estimate human emotional states by analysis of sleep stage. Thus, proposed analytic method (accurate and convenience for sleep stage estimation) could be understood to emotion and cognition of human easily.