Dual-Domain Experiment for EEG Emotion Recognition

Electroencephalography (EEG) is a pivotal tool for emotion recognition in computing and neuroscience fields. This study proposes a dual-method investigation by utilizing frequency-domain and time-domain EEG features. In the first experiment, EEG signals of 24 young adults were recorded when they listened to Kannada music had to evoke sad, relaxing, and enjoyable states. The data were band-pass filtered for 0.5–45 Hz which was trans-formed into frequency domain using Fast Fourier Transform (FFT), and features like band power, spectral centroid, spectral spread, spectral flux, and spectral flatness were extracted by utilizing the Power Spectral Density (PSD). Random Forest (RF) and KNN classifiers gained comparable accuracies of 73.33% (Channel 1) and 53.33% (Channel 2). In the second experiment, emotions such as funny, sad, and scary were classified while watching the advertisements by using timedomain statistical features like mean, variance, standard deviation, skewness, and kurtosis from the pre-processed EEG. The dataset was improved with subject-to-subject distance measures and clustering before classification by utilizing Multi-Layer Perceptron (MLP) neural network which achieved 83.33% accuracy on both Channel 1 and Channel 2. These results shows the distinct advantages of frequency-domain and time-domain analyses for culture-based emotion studies and consumer-neuroscience applications, respectively.