Evaluating the Effectiveness of Classification Algorithms for EEG Sentiment Analysis

Electroencephalogram (EEG) signals from the brain provide additional information about emotional states that we may be unable to convey verbally. Machine learning algorithms can effectively predict the emotion from brain waves. So, we design a research to evaluate the effectiveness of multiple machine learning techniques - Naive Bayes, Logistic Regression, XGBoost, SVM, Decision Tree, Random Forest, KNN, and deep learning models—CNN, LSTM, and Bi-LSTM for classifying sentiment from brain signals. In our experiment, the DEAP dataset is used as a collection of brain signals representing different human sentiments. The Fast Fourier transformation (FFT) which shifts the data to the frequency domain is used to extract features from the time series EEG data. Among all the algorithms CNN, KNN and Random Forest achieved the highest accuracy of 96.64%, 95.8%, and 95.28% respectively on the binary classification of valence. The results demonstrate that it is possible to attain accuracy comparable to or even outperform some of the deep learning models by combining appropriate feature extraction techniques (in this case FFT) with machine learning algorithms.