Multi-Output Regression for Integrated Prediction of Valence and Arousal in EEG-Based Emotion Recognition

Previous research in electroencephalogram (EEG)- based emotion recognition has assumed valence and arousal as independent dimensions and derived results using separate classification models for each. Our study aims to introduce multi- output regression to continuously predict the correlated valence...

Full description

Saved in:
Bibliographic Details
Main Authors: Choi, HyoSeon, Woo, ChaeEun, Kong, JiYun, Kim, Byung Hyung
Format: Conference Proceeding
Language:English
Subjects:
Arousal
Brain modeling
Brain-computer interfaces
EEG
Electroencephalography
Emotion recognition
Mirrors
Multi-output Regression
Predictive models
Valence
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Previous research in electroencephalogram (EEG)- based emotion recognition has assumed valence and arousal as independent dimensions and derived results using separate classification models for each. Our study aims to introduce multi- output regression to continuously predict the correlated valence and arousal. We compared the prediction performance of a total of three models: Single-output Regression, Multi-output Regression, and Multi-output Regression with a Chain structure. The analysis yielded the following results: 1) The performance of multi-output and single-output was similar, but the multi-output method is more useful as it can produce multiple outputs at once. 2) Predictions and actuals for valence and arousal were similar. Furthermore, considering that human emotions are not constant during certain intervals, it can provide insights close to the ground truth for understanding changes in emotional states.
ISSN:2572-7672
DOI:10.1109/BCI60775.2024.10480527