- Book Chapter
- 10.1007/978-981-19-4786-5_3
Mental Workload Classification Method Based on Transfer Component Analysis with Cross-Session EEG Data
- Aug 21, 2022
- Hongquan Qu + 2 more +2
Due to the non-stationary randomness of the EEG signals, which results in a trained classifier with very low generalization ability. In order to improve the cross-session classification accuracy of Mental Workload (MW), a domain adaptive classification method is studied based on Transfer Component Analysis (TCA) using EEG Independent Components (ICs) in this paper. First step is to transform the EEG data in source and target domains in a same way. The TCA algorithm is used to build the transformation ϕ in order to minimize their Maximum Mean Discrepancy (MMD) distance between the source domain and the target domain. Second step is to train a MW classifier using the SVM and the transformed EEG data in the source domain. The target domain data are used for test. The presented method is compared with Subspace Alignment (SA). EEG data in the source and target domains are measured 24 h apart. A-distance is calculated to reveal the distribution difference between the transformed source and the target domains. The results show that the classification accuracy of the proposed method can be improved by 15.11% compared with SA. The A-distance of TCA is significantly lower than that of SA.KeywordsEEGClassification of mental workloadIndependent Component Analysis (ICA)Transfer Component Analysis (TCA)Subspace Alignment (SA)
Read more