Summary
ChiMerge is a general, robust discretization algorithm
that uses the x2 statistic to determine interval similar-
ity/difference as it constructs intervals in a bottom-up
merging process. ChiMerge provides a useful, reliable
summarization of numeric attributes, determines the
number of intervals needed according to the character-
istics of the data, and empirical testing indicates sig-
nificant improvement over simple methods that do not
consider the class information when forming intervals.