Dynamic Re-ordering in Mining Top-k Productive Discriminative Patterns
Yoshitaka Kameya, K. Ito · 2017
Discriminative patterns are the patterns that distinguish transactions in two different classes, one of which is typically of our particular interest. They are also known under the names of emerging patterns, contrast patterns, subgroup descriptions, and so on. In order to reduce the search space for top-k productive discriminative patterns, this paper proposes to re-order sibling patterns dynamically according to their quality. It is formally shown that the "sub-patterns first" property, which makes it easy to test the productivity of patterns, still holds for a re-ordered enumeration tree. Moreover, in an extensive experiment, we observed that the proposed method shows a stable performance in various settings, and reduces the search space drastically for some burdensome situations. It is also found that the proposed algorithm works well as an anytime algorithm.