ExAnte: a preprocessing method for frequent-pattern mining

Abstract_Discovering frequent patterns in large datasets is one of the more pervasive data mining tasks. Albeit rooted in market basket analysis, frequent pattern mining can be adopted in many applications, and on data sources of different nature and structure; it also provides a basis for several other mining tasks, such as association rules, classification, and clustering. However, frequent pattern mining is inherently difficult, in that it handles typically too many input data, which typically yield too many patterns as a result - this is often an insuperable obstacle, both for performance limitations and for the impossibility to discern the interesting patterns from the many, mostly uninteresting, extracted ones. Preprocessing based on data reduction and userspecified constraints may be the solution to this problem: it may drive the mining process towards potentially interesting patterns, while enabling query optimizations at the same time. We show how this can be achieved on the basis of a simple yet powerful idea: combine constraints of different nature to the purpose of dramatically reducing the input database. The mining process after such preprocessing is strikingly optimized, both in terms of performance, and in capability of focussing on interesting patterns.