EEG Based Emotion Recognition Using Long Short Term Memory Network with Improved Rat Swarm Optimization Algorithm

Abstract

The automatic human Emotion Recognition (ER) based on Electroencephalography (EEG) signal has gained more attention among the researcher communities with a rapid growth of Human Computer Interaction (HCI). Most of the prior models have not focused on the context-information of the EEG signals. In this research manuscript, a novel automated model is implemented for improving ER using EEG signals. In the initial phase, the signals are acquired from an online database: Database for Emotion Analysis using Physiological Signal (DEAP). Then, the data denoising is carried-out by implementing Empirical Mode Decomposition (EMD) and Variational Mode Decomposition (VMD) filters. These filters aim in eliminating the artifacts and noises in the acquired raw EEG signals, and further, the feature extraction is carried-out utilizing 20 statistical features that extracts discriminative feature information from the decomposed EEG signals. In the last phase, the Long Short Term Memory network (LSTM) is used for human ER as arousal or valence. Additionally, the optimal hyper-parameters of the LSTM network are selected by proposing the Improved Rat Swarm Optimization Algorithm (IRSOA). As denoted in the resulting and discussion section, the IRSOA-LSTM network achieved a mean accuracy of 84.89%, sensitivity of 86.95%, specificity of 86%, precision of 83.68%, and f1-score of 85.28% on the DEAP database. The simulation outcomes state that the proposed IRSOA-LSTM network is better than the existing machine-learning models.