MUSIC Model based Neural Information Processing for Emotion Recognition from Multichannel EEG Signal

Emotion recognition from neuro-signal is a computationally challenging issue in the field of medical data science that has several interesting applications in mental state revelation. Usually, electroencephalogram (EEG) based neuro-signal is widely popular for its temporal resolution, portability, easy to use, and non-invasive features. Emotion recognition from multichannel EEG signal processing technique customarily depends on non-parametric frequency spectrum estimation. These methods often fail to achieve a significant assortment in frequency estimation of different emotional EEG signals due to the non-stationary behavior. In this work, MUSIC (Multiple Signal Classification), a parametric-based frequency spectrum estimation technique is proposed to extract features from multichannel EEG signals for emotional state classification. The proposed work utilized the SEED EEG dataset of three class emotional states for feature extraction and classified it with Multi-Layer Perceptron (MLP) network. The research analyzes the characteristics of the extracted MUSIC feature space through intensive visualization and compares the quality of the extracted feature space with conventional parametric model-based feature space. The average classification accuracy of the proposed method is 90%.