Person-Independent Emotion and Gender Prediction (EGP) System Using EEG Signals

Abstract

This paper presents a person-independent Emotion and Gender Prediction (EGP) system using Electroencephalography (EEG) brain signals. First, Short Time Fourier Transform (STFT) technique is implemented to get the time-frequency information for the selected electrode (Fz Electrode). Then, it is splitted into twenty sequential batches according to the electrode recorded time in seconds, and the maximum EEG activation voltage is located for every frequency level within each batch to create a 2D time-frequency extraction feature. Next, sparse auto encoder is applied to convert the distribution of the extracted feature into more compact and distinguished one instead. For system evaluation, Human-Computer Interaction) database (MAHNOB-HCI) public dataset with five-fold-cross validation classifier are used and implemented. In experiments, the proposed extracted feature improves the results of both emotion and gender prediction. For emotion prediction, the highest average accuracy is 97.07\%, 93.27% and 91.72\% for three, four and six emotions with Convolutional Neural Network (CNN) classifier, respectively. While, for gender prediction, experiments are tested related to neutral, amusement, happy, sad, and the mix of all these emotions, the highest average accuracy is obtained with CNN classifier in all emotion states (>95%) including the state of mixing all emotions together. As well as, the ability to distinguish between genders in case of mixing different emotions together is practically approved.