Feature Extraction from GAMEEMO data sets for Emotion Recognition

Abstract

Abstract— Feature extraction from the GAMEEMO dataset for emotion detection aims to identify boring, calm, frightening, and amusing emotional states. It is based on the isolation of significant patterns from raw EEG signals. The dataset consists of recordings of twenty-eight participants who engaged in four distinct computer games for five minutes each. Using a 14-channel Emotiv Epoc+ EEG instrument is crucial to extracting relevant information from the recorded brain signals.  Time-domain and frequency-domain analysis are common methods for feature extraction in emotion identification. The dynamics of brain activity may be represented from a time-domain perspective by computing the Hjorth parameters (activity, mobility, complexity), statistical measures (mean, variance, skewness, kurtosis), and signal differences (first and second difference). Band power ratios such as those for the Alpha, Beta, Delta, Gamma, and Theta bands are examples of frequency-domain characteristics that reveal information on power distribution across various frequency bands in the brainwaves. Furthermore, non-linear properties like fractal dimension and Lyapunov exponents, as well as entropy measurements like Tsallis, Renyi, and Shannon Entropy, may be used to describe complicated brain activity patterns. The characteristics that were retrieved are being used as crucial inputs by machine learning classifiers to distinguish emotional states with extremely high accuracy from EEG data included in the GAMEEMO dataset. MATLAB was used to generate the results that were reported in the study. All analyses and calculations were completed via MATLAB platform.