Systematic Literature Review for Emotion Recognition from EEG Signals

Researchers have recently become increasingly interested in recognizing emotions from electroencephalogram (EEG) signals and many studies utilizing different approaches have been conducted in this field. For the purposes of this work, we performed a systematic literature review including over 40 articles in order to identify the best set of methods for the emotion recognition problem. Our work collects information about the most commonly used datasets, electrodes, algorithms and EEG features, as well as methods of their extraction and selection. The number of recognized emotions was also extracted from each paper. In the analyzed articles, the SEED dataset turned out to be the most frequently used. The two most prevalent groups of electrodes were frontal and parietal. Evaluated papers suggest that alpha wavelets are the most beneficial band for feature extraction in emotion recognition. FFT, STFT, and DE appear to be the most popular feature extraction methods. The most prominent algorithms for feature selection among analyzed studies were classifier-dependent wrappers, such as the GA or SVM wrapper. In terms of predicted emotions, developed models in more than half of the papers were designed to predict three emotions. The predictive algorithms that were mostly used by researchers are neural networks or vector machine-based models.