On Feature Reduction with Application to Electroencephalograms.
Abstract
The report deals with the feature reduction problem in pattern classification, with application to electroencephalograph (EEG) signals. The problem under consideration is that of discriminating between two kinds of signals--spontaneous EEG and EEG driven by photic stimuli at the alpha frequency. Since an EEG record represents a large amount of data, efficient feature reduction methods are required to pick out a few features which are significant for discrimination purposes. The first two chapters are of an introductory nature describing statistical feature reduction methods given in the literature and some relevant facts about EEG signals. The third chapter develops a nonparametric feature reduction procedure based on a distance measure. The fourth chapter develops a random process model for the two kinds of EEG signals. The signal is essentially represented as a sinusoid at the alpha frequency with random amplitude and phase modulation. It is seen that the statistical properties predicted by the model agree closely with the observed results. In the fifth chapter, the model is employed for feature reduction and pattern classification. The model provides a four dimensional vector of sufficient statistics, which contains all the information necessary for discrimination purposes. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1970
- Accession Number
- AD0714174
Entities
People
- Karkal Pulkeri Sheshagiri Prabhu
Organizations
- Harvard University