Techniques of acceleration for association rule induction with pseudo-artificial life algorithm

Frequent pattern mining is an important problem in data mining. Generally, the number of potential rules grows rapidly as the size of the database increases. It is therefore hard for a user to extract association rules. To avoid such a difficulty, we propose a new method for association rule induction using a pseudo-artificial life approach. The proposed method is to decide whether there exists an item set which contains N or more items in two transactions. If it exists, a series of item sets that are contained in the part of the transactions will be recorded. The iteration of this step contributes to the extraction of association rules. It is not necessary to calculate a huge number of candidate rules. In an evaluation test, the authors compared the extracted association rules using our method with rules obtained with other algorithms such as the a priori algorithm. In an evaluation using a huge retail market basket data set, our method was found to be approximately 10 to 20 times faster than the a priori algorithm and its variants.