20432

An Efficient Sampling Approach for Mining all Association Rules in Large Databases

73

78

6

Mining association rules is an interesting problem in the field of knowledge discovery and data mining. In this paper a fast, new method for mining association rules is presented. In this new sampling approach, which utilizes FP-Growth to mine sample data, the candidate generation and test in the sample data is omitted because of the FP-Tree projection of sample data in the main memory. To overcome the main memory limitation in the new sampling method, a useful technique is proposed. Theoretical time consideration and empirical evaluation show that the new sampling approach is superior to the traditional Apriori-based sampling by orders of magnitude. Experimental evaluations on artificial and real-life datasets show that our approach, compared with a previously proposed sampling algorithm, is more efficient when the minimal support threshold is decreased and is also more stable as the size of the sample data increases.

1204469