Prediction Of Valence And Arousal Values Utilising Electroencephalogram (Eeg) Signals

Abstract

Emotions is a very important parameter in Brain-Computer Interface, as it helps the computer to accurately identify the user’s intentions with their commands. However, it is a difficult task to identify human emotions as there are many variations to the outward physiological expressions (facial expressions, tone of voice, etc). The proposed method aims to apply machine learning in processing electroencephalogram (EEG) signals, which is tied to brain activity, to identify human emotions by predicting Valence and Arousal values using 4 regressors: k-Nearest Neighbours (KNN), Support Vector Machine for Regression (SVR), Random Forest (RF), and Linear Regression (LR). The EEG signal used is the DEAP dataset with 4 features: power spectral density (PSD), wavelet energy (WP), wavelet entropy (WE), Hjorth’s mobility (H2) and complexity (H3) extracted from 4 frequency bands: theta (4 - 8 Hz), alpha (8 - 12 Hz), beta (12 - 30 Hz), and gamma (30 – 64 Hz) using Welch’s periodogram estimation, Discrete Wavelet Transform and Hjorth parameters. Cross-validation and feature standardization is then employed to process the features before being fitted into the machine learning algorithms. The results show that the best predictions are made by KNN and SVR with beta and gamma-based features.