IME: Efficient list-based method for incremental mining of maximal erasable patterns

Abstract

Erasable pattern mining can help factories facing a financial crisis increase productivity by identifying and eliminating unprofitable products. The Flag-GenMax-EI algorithm extracts Maximal Erasable Itemsets (MEIs); however, it does not support dynamic data. In practice, many applications create databases incrementally. Using the Flag-GenMax-EI algorithm to mine maximal erasable patterns from incremental databases is clearly very costly because it must be run each time. In this paper, an efficient method called IME is proposed for incremental mining of maximal erasable patterns. IMEI-List and IMEP-List are two new data structures introduced by the proposed method. These lists allow the algorithm to update all tree nodes without rescanning the updated database (original database + new database) and recreating the nodes. This is the first study of incremental mining of maximal erasable patterns. Extensive experimental results on dense and sparse incremental data show that the proposed algorithm improves scalability. It extracts MEIs much faster than the Flag-GenMax-EI algorithm in different modes of database update.