Mining maximal frequent itemsets for large scale transaction databases

We present a graph-based algorithm, MFIminer, for mining maximal frequent itemsets (MFI) from transaction databases. Our method is especially efficient in large transaction databases because the performance is not sensitive to the quantity of transactions. MFIminer adopts a directed association graph to guide the mining task efficiently. It uses the technique of depth-first traversal and complete graph checking to achieve reduction of searching time. Performance study shows that MFIminer outperforms minmax, an algorithm to find MFI, in both speed and scalability property.

1. Introduction

Mining frequent itemsets (F1) or patterns plays a fundamental and essential role in mining association rules, which is widely applied in many fields such as consumer market-basket analysis and inferring rules from gene expression data [1], [2]. Apriori and support-confidence framework, which is the basic of many algorithms generated recently for mining association rules, were originally proposed by Agrawal et al. [3], [4].