An approach for mining non-redundant sequential rules efficiently

Abstract

Mining sequential rules helps discover useful sequences in sequence databases quickly and efficiently. Most of the proposed algorithms, however, focus on generating all possible sequential rules. That will produce a lot of redundant rules, affecting efficient mining. In order to solve this problem, mining non-redundant sequential rules has thus been presented lately. However, the algorithms proposed for it depend on obtained patterns of the existing frequent pattern mining algorithms. That is several steps need to be done to organize the data structure of these patterns before being used for generating rules efficiently. This phase also takes a lot of time and memory usage. In this paper, we propose a technique to mine non-redundant rules from a sequence database directly. The proposed algorithm uses a compressed data structure and adopts a prefix tree in the mining process. Moreover, the proposed algorithm uses some pruning techniques to remove unpromising candidates early that show the efficiency of the algorithm in term of runtime and memory usage.