A new approach to generate frequent patterns from enterprise databases

Abstract

As data mining techniques are explored extensively, incorporating discovered knowledge into business leads to superior competitive advantages. Most techniques in mining association rules nowadays are designed to solve problems based on de-normalized transaction files. Namely, normalized transaction tables should be transformed before mining methods could be applied, and some previous works have pointed that such data transformation usually consumes a lot of resources. As a result, this study proposes a new method which incorporates mining algorithms with enterprise transaction databases directly. In addition, in most well-known mining algorithms, the minimum support threshold is used in deciding whether the pattern is frequent or not, and it is crucial to define an appropriate threshold before performing mining tasks. Since setting an appropriate threshold cannot be done intuitively by domain experts or users, they usually set the threshold through trial and error. Usually, while setting different minimum support thresholds, most existing algorithms re-perform all mining procedures. Consequently, it takes a lot of computations. Our new method explores such circumstances and provides ways to flexibly adjust support thresholds without re-doing the whole mining task. © Springer-Verlag Berlin Heidelberg 2005.