Experiments have been run on the closest matching strategy on the same different sized training data sets as in the previous experiments. We have used both the first (i.e., the number of matching conjuncts) and second methods (i.e., the number of unmatched conjuncts) to define the closest matching rule as listed in the previous section. The results are averaged over 25 trials. In each trial, the original data sets are split as described in the previous chapters. Then the new version of BruteDL, with the closest matching rule used before the default rule whenever an observation is not covered by any of the existing BruteDL rules, is run on 10%, 20%, ... 70% sized training sets.