Mining Frequent Patterns via Pattern Decomposition

• Candidates Generation and Test (Agrawal &Srikant, 1994; Heikki, Toivonen &Verkamo, 1994; Zaki et al., 1997): Starting at k=0, it first generates candidate k+1 itemsets from known frequent k itemsets and then counts the supports of the candidates to determine frequent k+1 itemsets that meet a minimum support requirement. • Sampling Technique (Toivonen, 1996): Uses a sampling method to select a random subset of a dataset for generating candidate itemsets and then tests these candidates to identify frequent patterns. In general, the accuracy of this approach is highly dependent on the characteristics of the dataset and the sampling technique that has been used. • Data Transformation: Transforms an original dataset to a new one that contains a smaller search space than the original dataset. FP-tree-based (Han, Pei & Yin, 2000) mining first builds a compressed data representation from a dataset, and then, mining tasks are performed on the FP-tree rather than on the dataset. It has performance improvements over Apriori (Agrawal &Srikant, 1994), since infrequent items do not appear on the FP-tree, and, thus, the FPtree has a smaller search space than the original dataset. However, FP-tree cannot reduce the search space further by using infrequent 2-item or longer itemsets.