A disagreement based co-active learning method for sleep stage classification

摘要

In this paper a co-active learning framework that uses two feature views for sleep stage classification is proposed. At the beginning of active learning, classifiers are trained on two separate feature views namely Empirical Mode Decomposition (EMD) and frequency domain based energy features which are obtained from different patients' labelled electroencephalography (EEG) data. These classifiers are updated on new patients' unlabeled data using a disagreement based co-active learning framework. Experimental results obtained on Sleep-EDF database show that the proposed co-active learning method outperforms single view based active learning and the proposed method boosts the classification accuracy to approximately 84.85% on average.