It frequently happens that the classes defined by the end user are not
the "natural" classes. For example, while the goal in classifying
English letters might be just to decide on one of 26 categories, if both
upper-case and lower-case letters will be encountered, it might make more
sense to treat them separately and to build a 52-category classifier, OR-ing
the outputs at the end. Of course, it is not always so obvious that there
are distinct subclasses.
Solution: Use a clustering procedure to
find subclasses in the data, and treat each subclass as a distinct class.
(We treat clustering in another section of this course.)
Back to Curved
On to Complex
Up to Limitations