Emotion Detection Research: A Systematic Review Focuses on Data Type, Classifier Algorithm, and Experimental Methods

Abstract

There is a lot of research being done on detecting human emotions. Emotion detection models are developed based on physiological data. With the development of low-cost wearable devices that measure human physiological data such as brain activity, heart rate, and skin conductivity, this research can be conducted in developing countries like Southeast Asia. However, as far as the author's research is concerned, a literature review has yet to be found on how this research on emotion detection was carried out in Southeast Asia. Therefore, this study aimed to conduct a systematic review of emotion detection research in Southeast Asia, focusing on the selection of physiological data, classification methods, and how the experiment was conducted according to the number of participants and duration. Using PRISMA guidelines, 22 SCOPUS-indexed journal articles and proceedings were reviewed. The review found that physiological data were dominated by brain activity data with the Muse Headband, followed by heart rate and skin conductivity collected with various wristbands, from around 5-31 participants, for 8 minutes to 7 weeks. Classification analysis applies machine learning, deep learning, and traditional statistics. The experiments were conducted primarily in sitting and standing positions, conditioned environments (for developing research), and unconditioned environments (applied research). This review concluded that future research opportunities exist regarding other data types, data labeling methods, and broader applications. These reviews will contribute to the enrichment of ideas and the development of emotion recognition research in Southeast Asian countries in the future.