Chapter Summary

This chapter studied BruteDL+, which is the modified version of BruteDL that stores all the homogeneous but not-best rules before the default rule if their Laplace accuracies are higher than that of the default rule. As the experimental results suggest:

• BruteDL+ performs better than the storing-all-rules BruteDL.
• BruteDL+ provides the same or slightly better performance than the original BruteDL in general. However, the differences are small and more experiments are required to confirm their significance. In addition, BruteDL+'s performance is terribly worse on one of the data sets, but it is likely that this could be remedied with some thought.

Therefore, BruteDL+ performs roughly the same as BruteDL does and is not as attractive as the system with Minimum OSR is. Nonetheless, the basic approach appears to hold some promise; if pursued, however, this is really best regarded as a modification of the criteria that BruteDL uses to construct, retain, and order rules; it is probably not best regarded as an attempt to employ a different default strategy.