EEG-based Negative Emotion Classification while Self Driving in a Simulated Environment

Abstract

The accuracy of the EEG signal analysis depends on the signal’s quality, preprocessing, features and classification algorithm. This study explores the Self-Assessment Manikin (SAM) and electroencephalography (EEG) with machine learning to classify vehicle driver emotions. Using SAM, emotional states such as relaxation, focus, fear, nervousness and surprise were quantified on a scale of 1 to 5 during experimental scenarios. EEG signals, captured via the Emotiv Epoc X, underwent preprocessing with a 6th-order bandpass filter and zero-phase distortion filtering to minimize artefacts and preserve signal integrity. Five time domain features and three frequency domain features are extracted for the classification. The mutual information (MI) used to reduce the features number and selected the features with the most significance only. A k-Nearest Neighbors (KNN) algorithm and Random Forest (RF) were applied to classify emotion and evaluating accuracy. The classification accuracy for KNN and RF are 98% and 93% respectively. These findings could lead to improvements in affective computing and driver monitoring systems. Compared to the past EEG emotion studies that rely on large feature sets, this research introduces a compact feature selection method based on mutual information to identify the most discriminative EEG features for negative emotions in autonomous driving simulation.