The way RGEN traverses the lattice of conditional attributes is unique as of today. The only similar method we have found is described in [MG94]. The traversal of RGEN leads to a broader set of rules than through the use of RSES and LERS.
Classification of unseen objects has yielded good results when using the combination RGEN and RCLASS with noisy data sets. The other types of data mining systems we have looked at also have noise handling capabilities, but this is often something added as an extra facility and not incorporated directly in the method.
The main problem with RGEN is its problems with execution speed. For
data sets larger than 2000 objects or more than 15 attributes, it becomes
impractical in use. This only reflects the fact that RGEN and RCLASS\
are research prototypes and by no means are intended for commercial
use. Our system is a general data mining systems and not
intended for specific use. It therefore has no way of competing
with domain specific applications for e.g. insurance, but will probably
both slower and more difficult to use in those cases. Still, the
combination of RGEN and RCLASS is applicable to a wide range of
problems and will create a broad set of useful default rules.