Analysing Effect of Database Grouping on Multi-Database Mining

In many applications we need to synthesize global patterns in multiple large databases, where the applications are independent of the characteristics of local patterns. Pipelined feedback technique (PFT) seems to be the most effective technique under the approach of local pattern analysis (LPA). The goal of this paper is to analyse the effect of database grouping on multi-database mining. For this purpose we design a database grouping algorithm. We introduce an approach of non-local pattern analysis (NLPA) by combining database grouping algorithm and pipelined feedback technique for multi-database mining. We propose to judge the effectiveness of non-local pattern analysis for multi-database mining. We conduct experiments on both real and synthetic databases. Experimental results show that the approach to non-local pattern analysis does not always improve the accuracy of mining global patterns in multiple databases. Index Terms — Local pattern analysis, Multi-database mining, Non-local pattern analysis, Pipelined feedback technique, Synthesis of patterns