Integrating Frequent Pattern Mining from Multiple Data Domains for Classification英文文献资料.pdf

2012 IEEE 28th International Conference on Data Engineering2012 IEEE 28th International Conference on Data Engineering Integrating Frequent Pattern Mining from Multiple Data Domains for Classication Dhaval Patel Wynne Hsu Mong Li Lee National University of Singapore Singapore {dhaval,whsu,leeml}@.sg Abstract—Many frequent pattern mining algorithms have been to mine useful frequent patterns for classication [10], [24], developed for categorical, numerical, time series, or interval [14], [5], [6]. data. However, little attention has been given to integrate these A straightforward method to discover heterogenous patterns algorithms so as to mine frequent patterns involving multiple data domains for classication. In this paper, we introduce the is to apply different frequent pattern mining algorithms for the notion of a heterogenous pattern that captures the associations different data domains, followed by an exhaustive combina- among different data domains. We propose a unied framework tions of the discovered patterns. A quick calculation reveals for mining multiple domains and design an iterative algorithm that this approach is computationally infeasible. A small called HTMiner. HTMiner discovers essential heterogenous pat- dataset with 10 categorical attributes, 20 numerical attributes, terns for classication and performs instance elimination. This insta