Comparison of t-test ranking with PCA and SEPCOR feature selectionfor wake and stage 1 sleep pattern recognition in multichannelelectroencephalograms

Padmashri, T K and Sriraam, N (2017) Comparison of t-test ranking with PCA and SEPCOR feature selectionfor wake and stage 1 sleep pattern recognition in multichannelelectroencephalograms. Biomedical Signal Processing and Control, 31 (31). pp. 499-512. ISSN 1746-8094

[img] PDF
1860.pdf - Published Version
Restricted to Registered users only

Download (2MB) | Request a copy


tFeature selection is critical for effective analysis of data and resource savings. In multi-dimensionaldatasets, feature selection methods mainly use filter based approach to obtain an optimal feature sub-space and wrapper methods to search for an optimal feature subset within this space. In the proposedstudy, two filter based statistical feature selection methods viz., statistical t-test ranking with principalcomponent analysis (PCA) and Separability & Correlation (SEPCOR) analysis are applied to identify pat-terns with high discrimination between wake and stage 1 sleep of a 8-channel (6 active +2 referenceelectrodes) electroencephalogram (EEG) sleep dataset. The feature set consists of 6-dimensional SpectralEntropy vectors computed over EEG epochs of one second duration. In the first method, spectral entropyfeature ranking is based on a t-test statistic that maximizes class separation between wakefulness/stage1sleep. Prior to classification, PCA is performed on the ranked and non-ranked feature subsets to studythe contribution of ranked channels on classifier performance. The second method uses SEPCOR anal-ysis to automatically select an optimal feature subset with low correlation among the chosen featuresand maximum separation between their class means. A correlation threshold is chosen heuristically insteps of 0.05 from 0.6 to 0.75 in order to select different subsets of features. The optimal feature sub-sets are evaluated using multi layered perceptron (MLP) network & k-nearest neighbor (k-NN) classifierswith 50% hold out cross validation. For ranked feature subsets N = 3, 4, 5, k-NN classifier outperformsMLP network with an increase in the number of principal components (pcs). Results indicate that thepcs of ranked channels enhance the performance of k-NN classifier whereas MLP network shows only amarginal improvement with ranking for number of channels, N ≤ 4. As the number of pcs is varied from2 to 4 in steps of one, there is an improvement of approximately 2% in the classification accuracies ofk-NN classifier with ranking as compared to their non-ranked counterparts. The MLP exhibits only 1%improvement with ranking for the same case with number of hidden neurons, N = 50. The k-NN classifierresponds with maximum accuracies of 96.43%, 95.7% and 94.10% (pc = 4, 3, 2 for no. of ranked channels,N = 4) as compared to 94.71%, 93.13% and 92% (pc = 4, 3, 2 non-ranked N = 4) respectively. The SEPCORresults show that with correlation threshold increasing from 0.6 to 0.75 in steps of 0.05, it automaticallyselects feature subsets of 2, 3, 4 and 5 which contribute to detection accuracies of 72.4%, 80%, 91.6% and92% with k-NN classifier and improved accuracies of 73%, 85%, 95.6% and 95.8% with MLP network (no. ofhidden neurons, N = 50) respectively. The SE feature ranking provides better classification results usingk-NN classifier than non-ranked cases whereas features obtained using SEPCOR analysis prove to be bet-ter discriminators with MLP network for the classification of wake/stage1 sleep data. The computationspeed is faster in k-NN classifier and independent of increase in value of k whereas MLP takes much morecomputation time for training based on the number of hidden neurons.

Item Type: Article
Uncontrolled Keywords: Electroencephalogram (EEG)Spectral entropyt-TestPrincipal component analysis (PCA)Separability & cor
Subjects: Engineering > MIT Manipal > Electronics and Communication
Depositing User: MIT Library
Date Deposited: 11 Jan 2017 11:32
Last Modified: 11 Jan 2017 11:32

Actions (login required)

View Item View Item