Abstract:
EEG (Electroencephalography) allows eliciting the mental state of the user, which in turn reveals the user emotion, which is an important factor in HMI (Human Machine Interaction). Researchers across the globe are developing new techniques to increase the EEG accuracy by using different signal processing, statistics, and machine learning techniques. In this work we discuss the most common techniques that can yield better results, along with discussing the common experiment steps to classify the emotion, starting from collecting the signal, and extracting the features and selecting the best features to classify the emotions, along with highlighting some standing problems in the field and potential growth areas.
In this work, we have identifed 10 different emotions based on Valance, Arousal and Dominance using five different models. EEG signals are collected and passed to the proposed models, the accuracy of the detection was ranging from 50% to 70%. Two sessions have been conducted per subject to collect the data for training and for testing the models. Evaluation has been conducted to assess the new model performance, the evaluation is measuring the performance of the model on external data that looks similar in shape to the data conducted in this study experiment. The self-assessment manikin (SAM) assessment technique has been used to tag the training data, but the SAM model comes with its own challenges, therefore, theoretical model has been proposed to accommodate for the SAM challenges as well as making the experiment easier to conduct, but making the model architecture more complex.
