A Comparison Between Two Association Rule Mining Techniques

Abstract

Data mining requires associated items to be mined from given transactions. Association rule mining (ARM) has gained importance in view of trend-prediction and decision-making process. Mostly ARM is implemented using variants of FP tree growth techniques. This paper introduces an improved technique for ARM – Improved Relative Dotted Sequence Path (IRDSP)-based ARM. The logic justifying why this technique saves time in comparison with classical technique – FP tree based-ARM has been explained. Projected technique saves time on two counts. On first count, the arrangement in descending order on the basis of occurrence frequency is not required. On second count, unlike FP tree, where each node and link is traversed many a times for generating frequent item set, this technique traverses them only once. The gain in time is because of time spent in formation of IRDSP graph. However as size of graph grows beyond a certain level, the proposed technique becomes comparable with other proposed approaches. Still the gain for small and moderate level set of transactions validates the effort.

Keywords: Association rule mining (ARM), frequent pattern (FP), improved relative dotted sequence path (IRDSP)