Distributed Incremental Data Mining from Very Large Databases: A Rough Multiset Approach

This paper presents a mechanism for developing distributed learners for learning production rules from massive, dynamic, and distributed databases. The task of distributed learning is formulated by the concept of multiset decision tables that is based on rough multisets and information multisystems, which are derived from the theory of rough sets. We use the concept of partition of boundary sets to represent and to combine distributed uncertain information. Learned rules are stored as multiset decision tables, which provide more compact representation, and they can be readily implemented using relational database technology.