Electroencephalogram Electrode Selection and Parallel Merged Recurrent Neural Network for Emotion Classification

The field of biomedical signal analysis for the development of Brain Computer Interface(BCI) has seen significant improvement recently through the development of modern Deep Learning Algorithms. The Electroencephalogram(EEG) signal is non-stationary, non-linear, and contains a lot of noise as a result of aberrations brought on by muscle action, blinking and poor electrode contact, making research in this field particularly difficult.The number of electrodes used to record EEG signals on non-invasive wearable devices raises the dimensionality and, consequently, the computing complexity of the EEG data. We propose the reduction of said dimensionality through extensive feature extraction and channel selection. The proposed model also employs Merged RNN with the help of Long Short Term Memory (LSTMs) applied on features from each selected channel.The Dataset for Emotion Analysis using Physiological Signals (DEAP), the research standard for emotion recognition, has been used to validate the model's applicability and accuracy.