Comparative analysis of dimensionality reduction techniques for EEG-based emotional state classification

Abstract

Objectives: The aim of this study is to evaluate the impact of various dimensionality reduction methods, including principal component analysis (PCA), Laplacian score, and Chi-square feature selection, on the classification performance of an electroencephalogram (EEG) dataset.

Methods: We applied dimensionality reduction techniques, including PCA, Laplacian score, and Chi-square feature selection, and assessed their impact on the classification performance of EEG data using linear regression, K-nearest neighbour (KNN), and Naive Bayes classifiers. The models were evaluated in terms of their classification accuracy and computational efficiency.

Results: Our findings suggest that all dimensionality reduction strategies generally improved or maintained classification accuracy while reducing the computational load. Notably, PCA and Autofeat techniques led to increased accuracy for the models.

Conclusions: The use of dimensionality reduction techniques can enhance EEG data classification by reducing computational demands without compromising accuracy. These results demonstrate the potential for these techniques to be applied in scenarios where both computational efficiency and high accuracy are desired. The code used in this study is available at https://github.com/movahedso/Emotion-analysis.

Keywords: Chi-square feature selection; Dimensionality reduction; K-nearest neighbour (KNN); Laplacian score; Naive Bayes classifiers; electroencephalogram (EEG); feature sets; linear regression; principal component analysis (PCA).