Abstract—An interstate business organization may have a large number of transactions in each of its branches operating at different locations. In such situations, for making effective head-quarter decisions, multi-database mining using local pattern analysis has been considered as an efficient strategy. During this process, individual data sources are mined and discovered patterns are forwarded to the head branch. To reap meaningful patterns from the large number of forwarded patterns, a comprehensive synthesizing process is a necessity. Earlier we had proposed a synthesizing model to synthesize global rules from high-frequent rules, in which weights are based on the transaction - population of sites. In this paper, we are proposing a synthesizing model for multi-level synthesis of local patterns on the basis of two, rule-selection measures -namely effective and nominal vote rates. Using these rule selection measures, synthesized patterns are classified into groupings of global, sub-global and local rules. With this, not only high-frequent rules but also frequent rules are taken care of and synthesized into appropriate set of sub-global rules. Examples and experimental results presented clearly establish the validity of the proposed model in meeting the requirements of multi-level rule synthesizing strategy.
Index Terms—Multi-Databases, Multi-level rule synthesis, Local pattern analysis, Effective vote rate, Nominal vote rate
1Thirunavukkarasu Ramkumar is with the Faculty of Computer Applications, AVC College of Engineering, Tamilnadu, India. (email: firstname.lastname@example.org)
2Rengaramanujam Srinivasan is with the School of Computer and Information sciences, BSA Crescent University, Tamilnadu, India. (email: email@example.com)
Cite: Thirunavukkarasu Ramkumar and Rengaramanujam Srinivasan, "Multi-Level Synthesis of Frequent Rules from Different Data-Sources," International Journal of Computer Theory and Engineering vol. 2, no. 2, pp. 195-204, 2010.