Sampling large database for association rules without candidate generation

In this study a new and fast approach is presented for mining association rule. It is a hybrid approach based on sampling algorithm which uses FP-Growth to find frequent itemset in sample data. This method neglect candidate generation and test due to FP-Tree projection of sample data. For the main memory limitation problem i.e. loading sample data and representing FPTree structure a useful technique is proposed. Both theoretical and experimental evaluations show that the approach is faster than Apriori-based sampling by orders of magnitude. In addition experimental evaluation shows that the approach is more efficient when the minimum support threshold is reduced.

کلیدواژه ها:

Data mining ، Associations Rule mining ، Apriori ، Sampling ، FP-Growth