Abstract
Effective data analysis using multiple databases requires highly accurate patterns. As the local pattern analysis might extract patterns of low quality from multiple databases, it becomes necessary to improve mining multiple databases. In this chapter, we present an idea of multi-database mining by making use of local pattern analysis. We elaborate on the existing specialized and generalized techniques which are used for mining multiple large databases. In the sequel, we discuss a certain generalized technique, referred to as a pipelined feedback model, which is of particular relevance for mining multiple large databases. It significantly improves the quality of the synthesized global patterns. We define two types of error occurring in multi-database mining techniques. Experimental results are provided and they are reported for both real-world and synthetic databases. They help us assess the effectiveness of the pipelined feedback model.