Abstract
This chapter focuses on the use of electroencephalogram (EEG) signals to classify human emotions by performing both feature extraction and feature selection for each channel of an Emotiv Epoc+ headset. For feature extraction, we use Shannon entropy, log energy entropy, differential asymmetry, and rational asymmetry. For feature selection, we only use one method, that is, a minimum redundancy and maximum relevance algorithm. The features for each channel give higher emotion classification accuracy (CA) compared to multi-channel features (MCFs) because they give more information per channel. We classify five emotions, namely, amused, sad, neutral, scared, and anger, using two supervised machine-learning algorithms, a random forest (RF) and a long short-term recurrent neural network (LSTM RNN). We perform a 14-channel EEG brain signal experiment with six subjects, four males and two females, aged 19-35, using a 12 min video stimulus to elicit the desired five human emotions. We process the signal data in terms of channel-wise features (CWFs) and compare them against the MCFs. We achieve a maximum CA for the LSTM RNN of 83.80% (CWF) and 36.00% (MCF), and for the RF of 54.90% (CWF) and 39.90% (MCF). Next, we repeat the same process to verify our proposed approach using a publicly available human emotion dataset, SEED-IV. In this dataset, we achieve a maximum CA for the LSTM RNN of 99.80% (CWF and MCF), and for the RF of 48.0% (CWF) and 46.1% (MCF). The results show that using channel-wise signal data processing retained feature information per channel necessary for achieving better accuracy. We conclude that this approach can be used in healthcare facilities to more accurately and quickly diagnose the emotional status of patients using their biophysical signals, and thereby deter any health problems that undetected emotions would have caused.
Original language | English |
---|---|
Title of host publication | Affective Computing in Healthcare |
Subtitle of host publication | Applications based on biosignals and artificial intelligence |
Publisher | Institute of Physics Publishing |
ISBN (Electronic) | 9780750351829 |
ISBN (Print) | 9780750351805 |
DOIs | |
Publication status | Published - Aug 3 2023 |
All Science Journal Classification (ASJC) codes
- General Engineering
- General Computer Science
- General Biochemistry,Genetics and Molecular Biology