Search

The most crucial feature of human computer interaction is computers and computer-based applications to infer the emotional states of humans or others human agents based on covert and/or overt signals of those emotional states. In emotion recognition, bio-signals reflect sequences of neural activity induced by emotional events and also, have many technical advantages. The aim of this study is to classify six emotions (joy, sadness, anger, fear, surprise, and neutral) that human have often experienced in real life from multi-channel bio-signals using machine learning algorithms. We have measured physiological responses of three-hundred participants for acquisition of bio-signals such as electrodermal activity, electrocardiograph, skin temperature, and photoplethysmograph during six emotions induction. Also, for emotion classification, we have extracted eighteen features from the signals and performed emotion classification using five algorithms, linear discriminant analysis, Naive Bayes, classification and regression tree, self-organization map and support vector machine. The used algorithms were evaluated by only training, 10-fold cross-validation and repeated random sub-sampling validation. We have obtained recognition accuracy from 42.4 to 100% for only training and 39.2 to 53.9% for testing. Also, the result for testing showed that an accuracy of emotion recognition by Naive Bayes and linear discriminant analysis were highest (53.9%, 52.7%) and was lowest by support vector machine (39.2%). This means that Naive Bayes is the best emotion recognition algorithm for basic emotions. To apply to real system, we have to discuss in the view point of testing and this means that it needs to apply various methodologies for the accuracy improvement of emotion recognition in the future analysis.

Emotions are complex processes involving multiple response channels, including physiological systems, facial expressions and voices. Bio-signals reflect sequences of neural activity, which result in changes in autonomic and neuroendocrine systems induced by emotional events. Therefore in human-computer interaction researches, one of the most current interesting topics in emotion recognition is to recognize human's feeling using bio-signals. The aim of this study is to classify emotions (joy, sadness, anger, fear, surprise, and neutral) that human have often experienced in real life from multichannel bio-signals using machine learning algorithms. We have measured physiological responses of three-hundred participants for acquisition of bio-signals such as electrodermal activity, electrocardiograph, skin temperature, and photoplethysmo-graph during six emotions induction. Also, for emotion classification, we have extracted eighteen features from the signals and performed emotion classification using four algorithms, linear discriminant analysis, Naive Bayes, classification and regression tree and support vector machine. The used algorithms were evaluated by only training, 10-fold cross-validation and repeated random sub-sampling validation. We have obtained recognition accuracy from 56.4 to 100% for only training and 39.2 to 53.9% for testing. Also, the result for testing showed that an accuracy of emotion recognition by Naive Bayes was highest (53.9%) and lowest by support vector machine (39.2%). This means that Naive Bayes is the best emotion recognition algorithm for basic emotions. This result can be helpful to provide the basis for the emotion recognition technique in human-computer interaction.

This study examined a relationship between anxiety scale score and two types of EEG asymmetry: resting state and fear-induced state. 30 adults who have no psychiatric history were participated, and their EEGs were recorded while resting and watching emotion-inducing film clips respectively. The result was obtained that there is no significant difference in resting state EEG asymmetry between high and low anxiety, but there is a meaningful difference in fear-induced frontal EEG asymmetry between two anxiety levels. The group with high anxiety level shows greater EEG asymmetry. We discussed the way how this result could be applied to anxiety treatment.

It is difficult to classify anger, fear, and surprise emotions with autonomic nervous system response patterns, because these three emotions show similar levels of valence and arousal dimensions. The purpose of this study was to classify three emotions by using EEG signals. Linear discriminant analysis (LDA) using three types of EEG characteristics showed that the mean recognition accuracy was 66.3%. These findings reveal that three emotions were successfully able to be classified based on EEG signals.

The purpose of this study is to identify optimal algorithm for emotion classification which classify seven different emotional states (happiness, sadness, anger, fear, disgust, surprise, and stress) using physiological features. Skin temperature, photoplethysmography, electrodermal activity and electrocardiogram are recorded and analyzed as physiological signals. The emotion stimuli used to induce a participant's emotion are evaluated for their suitability and effectiveness. For classification problems of seven emotions, the design involves two main phases. At the first phase, Particle Swarm Optimization selects P % of patterns to be treated as prototypes of seven emotional categories. At the second phase, the PSO is instrumental in the formation of a core set of features that constitute a collection of the most meaningful and highly discriminative elements of the original feature space. The study offers a complete algorithmic framework and demonstrates the effectiveness of the approach for a collection of selected data sets.

Recently, the one of main topic of emotion recognition or classification research is to recognize human's feeling or emotion using various physiological signals. It is one of the core processes to implement emotional intelligence in human computer interaction (HCI) research. The purpose of this study was to identify the best algorithm being able to discriminate negative emotions, such as sadness, fear, surprise, and stress using physiological features. Electrodermal activity (EDA), electrocardiogram (ECG), skin temperature (SKT), and photoplethysmography (PPG) are recorded and analyzed as physiological signals. And emotional stimuli used in this study are audio-visual film clips which have examined for their appropriateness and effectiveness through preliminary experiment. For classification of negative emotions, five machine learning algorithms, i.e., LDF, CART, SOM, Naive Bayes and SVM are used. Result of emotion classification shows that an accuracy of emotion classification by SVM (100.0%) was the highest and by LDA (50.7%) was the lowest. CART showed emotion classification accuracy of 84.0%, SOM was 51.2% and Naive Bayes was 76.2%. This can be helpful to provide the basis for the emotion recognition technique in HCI.