This is the first sub-task in our classification procedure; here we generate a set of candidate composite features. For this
sub-task we first transform the dataset so that each example is represented as a set of integers, as described in Section 3.
We then run a generic frequent itemset discovery algorithm on this dataset assuming each example as a transaction
and each attribute value in the example as an item. The Frequent Itemset Discovery Algorithm, henceforth referred as
FIDA, returns a list of itemsets which occur frequently in the dataset. Each itemset represents a composite feature that
is a conjunction of all the attribute values (items) making up that itemset. In our procedure we use LPMiner [SK01] as
our FIDA.