Multiclass Classification of EEG Motor Imagery Signals Based on Transfer Learning

Abstract

Multiclass Classification of EEG signal is essential for brain computer interface (BCI) applications but extremely time consuming. We proposed a subject-weighted adaptive transfer learning method in conjunction with MLP and CNN classifiers for fast classification of multiclass EEG dataset.Analytic results show that CNN generally outperforms MLP in this multi-class classification. The use of transfer learning is efficient for building the predictive model without decreasing the accuracy and 2D CNN is more robust to between-subject variabilities.