Natural Human Emotion Recognition Based on Various Mixed Reality(MR) Games and Electroencephalography (EEG) Signals
Abu Saleh Musa Miah, Jungpil Shin, Md. Minhajul Islam, Abdullah Abdullah, Md. Khademul Islam Molla
Abstract
We estimated willing and natural emotions while playing Mixed reality (MR) games. We have shown the performance accuracy of the labeling with game type and self-assessment EEG data. This study is conducted to improve the Virtual reality (VR) and MR world to be more realistic and suitable to the needs. We have used the GAMEEMO dataset to evaluate our proposed method. First, participants took subjective electroencephalography (EEG) signals while playing the games, then took their self-assessments for labeling EEG signals based on Game rating. There were four categories of Games Boring (G1), Calm (G2), Horror(G3), and Funny(G4) in the dataset. We labeled the collected dataset in two ways: the dataset with the type of games and with the Self-assessment manikin. After that, we calculated mean and standard deviation (STD) from both types of EEG data. Based on the feature vector, we applied a machine-learning algorithm to classify the emotion of the subject for the game based on EEG data. Specifically, we have employed multinomial logistic regression (MLR), support vector machine (SVM), K-nearest neighbor (KNN) machine learning algorithm with cross-validation. With the method, we achieved 99.80% of accuracy for SVM with STD and 99.00% for the MLR with the mean method. This accuracy is 3.91‒4.92% better than that of the existing work, which shows an improved understanding of MR and human emotion and helps understand better human emotion and the area of VR and MR, estimated willing emotion. Based on the study result, a better life for the next generation can be provided. Finally, this study is the first step to prevent negative MR content from spreading out.