Semantic video classification and feature subset selection under context and concept uncertainty