Frequent pattern mining is an unsupervised learning task — informally, "bump hunting" — that searches a remarkably flat distribution over an enormous outcome space for the subsets of items that co-occur more often than a stated minimum support threshold, where support is defined as the fraction of transactions containing an itemset. From frequent itemsets one derives association rules, whose usefulness is quantified by support, confidence (the conditional probability of the consequent given the antecedent), and lift (the ratio, or difference, between that conditional probability and the unconditional probability of the consequent). The topic belongs to data mining over transaction databases, and its key theoretical refinements are the closed and maximal itemset condensations, which shrink the output while preserving different amounts of recoverable information.
Frequent pattern mining is an unsupervised learning task — informally, "bump hunting" — that searches a remarkably flat distribution over an enormous outcome space for the subsets of items that co-oc…