We are considering a number of improvements and extensions to the CDW
algorithms. First, the CDW weighting algorithm could be extended to process
numeric features in addition to symbolic ones. The most straightforward way
to do this is to partition numeric features into histogram buckets. However,
this discards some of the information present in the numeric values. A better
extension would take into account the continuous nature of numeric features
while preserving the paradigm that the weight of a feature should be based
directly on its usefulness in distinguishing classes