FEATURE-SELECTION USING A PROXIMITY-INDEX OPTIMIZATION MODEL

One of the recurring issues in pattern recognition problems has been f eature extraction and selection. This paper addresses this issue from a different perspective. Without assuming any particular classificatio n algorithm, it first suggests that one extract as much information as conveniently possible in several pattern-information domains. This pa per later suggests applying the proposed Proximity-Index method, to se lect a significantly smaller, yet optimal feature subset. This method is formally described and is successfully applied to a waveform classi fication problem. The features selected by the algorithm are used to c lassify ten signal classes and produce a very encouraging recognition performance of 87.00% on 200 samples. This method is computationally i nexpensive and particularly useful for large data set problems.