Knowledge Hiding in Databases for concept-based data mining algorithms

摘要

One of the limitations of the techniques that have been developed as apart of the Knowledge Hiding in Databases (KHD) methodology is that they are not applicable to a general class of data mining algorithms. In this paper, we present a formal characterization of the KHD process for a general class of data mining algorithms, that we call concept-based. This particular class of mining algorithms includes decision-region based classification algorithms, association algorithms, negative association algorithms, and exception rule mining algorithms. All of these algorithms have the common feature that the patterns generated by them can be represented using Bacchus probability logic. Based on our concept of a pattern, each step of the KHD process is able to treat concept-based algorithms in a unified manner.