difference between the Equal Width discretization used as
a density estimator and a supervised discretization method, we display in figure 1
the density function of a continuous attribute with two target classes. A histogram
with a correct bin number properly identifies the three density peaks, but it may
mixture the two target classes in the middle peak. On the opposite, a supervised
discretization method perfectly separates the target classes by building two intervals,
but it is not sensitive to the underlying density. The objective of a good stacked
histogram is to consider the target classes in the visualization of the density.