EEG-based Emotion Recognition using DMCCA, ICA, CCA, PCA, and Deep Learning

EEG-based emotion classification is a highly sought-after application due to its relevance in healthcare, human-computer interfaces, as well as brain-computer interfaces. In this work, a new approach towards EEG-based emotion classification is presented with state-ofthe- art feature extraction methods in terms of Dependent Multiple Canonical Correlation Analysis (DMCCA), Independent Component Analysis (ICA), as well as Canonical Component Analysis (CCA). Between these methods, DMCCA outperformed the others with a great improvement in feature representation as well as classification. The approach employs Discrete Wavelet Transform (DWT) as a time-frequency decomposition followed by dimensionality reduction through Principal Component Analysis (PCA), which helps in maintaining key information. A deep learning-based Convolutional Neural Network (CNN) is adopted as a classification strategy. The experimental results depict that DMCCA outperformed all feature extraction methods with a classification rate as high as 97% with minimal classification error as confirmed by a confusion matrix. The work positions DMCCA as a highly effective strategy towards EEG-based emotion classification over traditional methods that are deep learningbased.