BROWSE
Detail
Conference Paper
ECG-based Emotion Classification using Optimized Machine Learning Techniques
Cited - time in 
Share 
Share
- Authors
- Gague Kim, Seungeun Jung, Jjyoun Lim, Kyoung Ju Noh, Hyuntae Jeong
- Issue Date
- 2019-12
- Citation
- International Symposium on Advanced Intelligent Systems (ISIS) 2019 / International Conference on Biometrics and Kansei Engineering (ICBAKE) 2019, pp.500-507
- Language
- English
- Type
- Conference Paper
- Abstract
- Recently, there has been a growing interest in research to recognize and process human emotions for natural interaction between humans and machines. Emotions can be analyzed along with the user’s actions and situations to grasp the user’s psychological states as a whole and furthermore, emotional interaction between humans and machines can be made possible by enabling the operation of smart devices based on emotions. This study deals with classifying two or four emotional states from ECG (Electrocardiography) signals. We extract HRV (heart rate variability) features form ECG and use KNN (k-Nearest Neighbor) algorithm for classification. Here, KNN input are feature transformed prior to classification and the transformation coefficient is determined by using the GA algorithm to maximize the classification performance. Based on the emotional states obtained through two independent emotional classifiers, we get four class emotional state through the decision tree. This two-layered emotional classifier is consistent with Russell’s emotional model. The proposed method has achieved 75.1% and 68.3% accuracy respectively in classifying two classes of emotions that is valence positive/negative and arousal high/low. 47.8% accuracy was achieved in four classes of emotional classifications. These results confirm the validity of the proposed method compared with the single layered method with 46.1% classification accuracy. In addition, the use of single physiological sensor is generally likely to be used for emotional recognition in daily life. Therefore, our emotional classification system, which uses only single ECG sensor shows that it can be fully feasible in applications that recognize users’ emotions during daily life.
- KSP Keywords
- Classification Performance, Classification system, Decision Tree(DT), Emotion classification, Emotional Recognition, Emotional classification, Emotional interaction, Emotional states, GA algorithm, Heart rate variability, Human Emotions