In general, multiple attributes need to be considered for classification. Then,
finding p(d|c j ) exactly is difficult, since it requires the distribution of instances
of c j , across all combinations of values for the attributes used for classification.
The number of such combinations (for example of income buckets, with degree
values and other attributes) can be very large.With a limited training set used to
find the distribution, most combinations would not have even a single training
set matching them, leading to incorrect classification decisions.