Feature Extraction and Classification Results for the Batchelor-Hand-Brumfitt Artificial Data Base,
Abstract
A study is presented on the performance of feature extractors and classifiers on the 10-dimensional artificial data base provided by B. G. Batchelor, D. J. Hand, and P. G. Brumfitt of the Department of Electronics, University of Southampton, Southampton. On feature extraction, specifically three techniques for the extraction of the best one dimensional subspace were examined on the above data: the single linear feature (SLF) extraction algorithm developed at Rice; the single feature (SFB) corresponding to the maximum Bhattacharyya distance between pattern classes; and the optimal Karhunen-Loeve (KLF) subspace. Classification performance on the test set was 73.2% for SLF 66.2% for SFB and 49.7% for KLF. Classification in subspace of dimension greater than one was also investigated. In the entire 10-dimensional space, a classification performance of 80.8% was attained using maximum likelihood classifier.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1976
- Accession Number
- ADA032897
Entities
People
- K. C. Pau
- R. J. P. De Figueiredo
- S. A. Starks
Organizations
- Rice University