Determination of Emotional States from Electroencephalogram (EEG) Data Using Machine Learning

Abstract

Brain functions can be analysed by recording electrical activity, which can be done through methods such as electroencephalography (EEG). In the present study, we focused on the identification of three emotional states using EEG signals: happiness, sadness and anger. We used data from a study published on OpenNeuro [1] and analysed it in order to decode emotional states from EEG signals. We first processed the data into their distinct frequency bands and computed the power spectral density, before extracting the spectral power of five frequency bands from the signal. Subsequently, we used a support vector machine (SVM) algorithm to classify the signals based on their corresponding emotional states. After including all frequency bands as features to train the SVM algorithm, we used 5-fold cross validation to determine the accuracy of our model which was 73.99% and significantly higher than chance-level accuracy of 33.33%. After comparing different frequency band combinations as features, using the delta, gamma, beta and theta frequency bands resulted in the highest accuracy, of 74.25%, for our SVM algorithm in classifying the emotional states of happiness, sadness and anger. We concluded that using all frequency bands as features does not necessarily result in the highest possible accuracy rate as some frequency bands lower the accuracy of the machine learning classifier.