Chapter Summary

After running experiments on the default strategy of using a closest-matching rule before using the default rule, we realized:

• Although the closest matching rule provides roughly the same accuracy on many data sets, it performs badly on some data sets.
• The default rule included in BruteDL's output actually has a hidden non-empty condition that contains the negation of all the previous rules' conditions. More generally, all rules (except the first) have implied conditions. A more sophisticated notion of similarity might take these implied conditions into account.

The idea of using the closest matching rule may hold some promise, but our current implementation does not consider some important subtleties.