Recursive Identification Based on OS-ELM for Emotion Recognition and Prediction of Difficulties in Video Game

Hayfa BLAIECH, Noureddine LIOUANE

Abstract

The aim of this paper is to recognize human emotions from physiological electroencephalography (EEG) signal. Indeed, this paper demonstrates that neural networks in conjunction with recursive least squares can be used sprucely for the identification of emotional states. In this regards, this report uses online sequential extreme learning machine to conceive the present emotion recognition system. To validate the efficiency of this machine learning, the performance of various popular feature extraction methods with the Database for Emotion Analysis using Physiological Signals DEAP dataset, and two newly developed EEG datasets are systematically evaluated for this study. The first emotional experience was to conceive an emotional dataset containing the samples EEG of six selected emotion states (neutrality, amusement, surprise, compassion, attraction, and disgust), and the second one was to predict the difficulties encountered during playing a hard game. This has the goal to detect the concentration and anger emotional states. The purpose of this paper is to improve the accuracy of emotion recognition based on different emotional states. The features are extracted from the delta, theta, alpha and beta bands. Based on the analysis of the identification system, the most significant features are extracted for each emotional state. The selected ones are then utilized in the emotion classification system. The present recognition system has 80% accuracy when using the DEAP database, 74.07% accuracy when using the first database, and obtains 61.11% accuracy when predicting difficulties in video games.

Keywords

Recursive identification, OS-ELM, Emotion Recognition, EEG signals, Video games, Feature extraction.

View full article