Combining: an alternative for choosing between reject-first andaccept-first classifiers

摘要

In the general framework of pattern classification two rejectstrategies are identified. These classifiers are defined as a couple oflabeling and hardening functions. In this paper the first stage ofclassifiers belonging to each strategy are used in turn and theirresults are combined with a Dempster-Shafer model to classify or rejectpatterns. Results on artificial data are provided