EEG Based Emotion Recognition Using Deep CNN Classifier and Hybrid Feature Selection Algorithm

Abstract

Electroencephalogram (EEG) based emotional evaluation has achieved excellent outcomes in medicine, security, and interaction between humans and computers. Especially compared with traditional signal processing and Machine Learning (ML) based applications, Deep Learning (DL) based techniques have recently dramatically increased the classification precision. Due to its sufficient spatial accuracy and enhanced temporal resolution, EEG signals typically represent emotional states. It is essential to consider that identifying emotions based on EEG signals relies on the efficacy of three processes: extracting features, selecting features, and classifying the feelings. Therefore, this work proposes a computerized approach for recognizing emotions from EEG signals. High Pass Infinite Impulse Response with Zero-Filtering (HPIIRZ) approach is used to reduce artifacts in EEG signals. Following this, the frequency and spectral features are extracted using Power Spectral Density (PSD), from which the optimal features are selected by a hybrid Improved Artificial Bee Colony algorithm-Particle Swarm Optimization (IABC-PSO). Deep Convolutional Neural Networks (DCNNs) are then used for classifying emotional states at the classification stage. An evaluation model is developed using the Python platform to evaluate the performance of the proposed model, including accuracy, specificity, and sensitivity. The outcomes demonstrate that the proposed method is more efficient; the DCNN-based method achieves a higher accuracy of 95.80%.