Tools of Analysis for Pattern Recognition.
GEORGIA UNIV ATHENS DEPT OF STATISTICS AND COMPUTER SCIENCE
Pagination or Media Count:
In the study, pattern recognition is considered to be a problem in data analysis. Techniques for determining the proximity of sample patterns to calibration patterns are developed and discussed. A major portion of the study is concerned with the problem of scaling categorized data, in which states of the response variables are measured on a nominal scale. A technique for scaling such variables, which assigns scale values to states of a response variable so as to produce maximum discrimination between levels of a selected criterion variable, is developed and discussed. It is shown that an exact test for group differences, using response variables previously scaled to provide maximum discrimination between the groups, exists and is based on the maximum characteristic root distribution. Indices for determining the quality of scale values assigned by a particular scaling technique are suggested. Modified author abstract
- Computer Programming and Software