Discriminant feature level fusion based learning for automatic staging of EEG signals

Abstract

Wide-scale information embedding is a prerequisite to enhance the performance as well as the reliability of decision-making algorithms for viable implementation. Feature fusion technology significantly helps to incorporate such information to provide promising algorithm performance. In this Letter, a fusion-based model with the aid of discriminant correlation analysis to classify electroencephalogram signals is proposed. Sets of multiple feature matrices are generated from signals in both time and wavelet domains for study-specific classes, which are further decomposed to derive a set of sub-multi-view features followed by optimisation to extract statistical features. Features are concatenated using feature fusion technique to derive low order discriminant features. Besides, the analysis of variance was also performed to validate the analysis. The statistically significant features are evaluated for the effective model performance. Experimental results manifest that the proposed feature fusion based algorithm is superior to many state-of-the-art methods and thus promote real-time implementation.