Exploring principal component analysis and convolutional neural network (PCA-CNN) based architecture for enhanced emotion classification in EEG signal extraction

The current work on emotion recognition considering EEG signals has attracted significant attention due to its advantages in fields like mental health, neural computing, and human-computer interaction. Nonetheless, accurately classifying emotions using EEG signals poses a challenge, mainly due to issues such as noise, artefacts, and high dimensionality found within EEG data. Moreover, the current problem lies in extracting meaningful emotional features while efficiently addressing the above-described challenges to improve classification accuracy. Hence, this work proposed an approach for solving the challenges in emotion classification by presenting a model that integrates Principal Component Analysis (PCA) and Convolutional Neural Network (CNN). The PCA was used for removing noise and for dimensionality reduction. In contrast, the CNN was used for extracting emotion-related features and classifying emotions, mainly the two classes, arousal and valence. The proposed PCA-CNN model was evaluated considering the DEAP dataset, where the PCA-CNN model achieved 93.58% accuracy for arousal and 92.75% for valence for subject-dependent and 81.40% accuracy for arousal and 79.84% for valence for subject-independent. The findings also show that the PCA-CNN model provided better results in handling noise and reducing data complexity, ensuring high classification performance. The novelty of the PCA-CNN model is that PCA was used for preprocessing the EEG data, and CNN was used for feature optimization, which created a more efficient emotion classifier.