Hyun-il Kim, Woong-sik Shin, Choon-sung Nam, Journal of Internet Computing and Services, Vol. 26, No. 6, pp. 51-62, Dec. 2025
10.7472/jksii.2025.26.6.51, Full Text:  HTML
Keywords: EEG, feature engineering, Pearson Correlation Coefficient, wavelet transform, OVR-CSP, Olfactory Classification

Abstract

While multivariate feature extraction methods, such as the Wavelet-Spatial Domain Feature (WSDF), demonstrate high classification performance in olfactory electroencephalogram (EEG) analysis, their reliance on all available channels results in significant computational costs, posing a major challenge for real-time applications. To address this issue, this study proposes a novel pipeline that integrates a Pearson correlation-based channel selection method as a preprocessing step to the WSDF algorithm. The objective is to achieve an optimal balance between computational efficiency and classification accuracy. The proposed technique systematically identifies and eliminates redundant channels by analyzing inter-channel correlations, thereby efficiently selecting an optimal subset of channels crucial for olfactory information processing. Experiments conducted on a public olfactory EEG dataset demonstrated that the proposed method reduces the feature extraction time by up to 58%. Notably, this significant gain in efficiency was achieved with no loss in classification accuracy; in fact, a slight improvement in accuracy was observed under certain conditions.

Statistics
Cite this article