Feature selection using a proximity-index optimization model
Pattern Recognition Letters
A Comparison of Seven Techniques for Choosing Subsets of Pattern Recognition Properties
IEEE Transactions on Computers
IEEE Transactions on Pattern Analysis and Machine Intelligence
Patterns in pattern recognition: 1968-1974
IEEE Transactions on Information Theory
Hi-index | 0.98 |
The proximity feature selection model developed by the authors is enhanced to consider a multicategory classification problem. The proximity indices between intra- and inter-categories are used to establish the similarity and dissimilarity criteria. Using these criteria, a bicriteria Boolean linear programming model is developed to select an optimal set of discriminatory parameters to solve a multicategory classification problem. This method, thus, addresses the recurring issue of parameter selection in pattern recognition problems, and suggests the designers of classification systems to consider the problem from a different perspective. The paper suggests that one shall extract as much information as conveniently possible in pattern-information domain, and use the proposed model to select a significantly smaller, yet optimal parameter subset for classification. This method is formally described and is successfully applied to an electroencephalogram (EEG) waveform classification problem. The parameters selected by the algorithm are used to classify three EEG signal classes, and produce a very encouraging recognition performance of over 86% on 300 samples from three classes. This method is computationally inexpensive and particularly useful for large data set problems.