Decision Boundary Analysis Feature Selection for Breast Cancer Diagnosis

摘要

The general pattern recognition problem always involves the extraction offeatures to be used in pattern classification. There are no theoretical limitations to the number of features which can be obtained for a given pattern recognition problem. This research will develop a correlation procedure for screening a large feature set without the use of a trained classifier. The results will be compared to established saliency metrics such as the Fisher ratio and derivative-based techniques such as Ruck's saliency.