The Effect of Correction Factor in Synthesizing Global Rules in a Multi-Database Mining Scenario

Abstract

Recently, multi-database mining using local patternanalysis has been identified as an efficient strategy for miningmultiple data sources of an interstate business organization.Using this approach, frequent patterns from the individualsites are synthesized and forwarded to the central head.Various synthesizing models [5,7] have been proposed to formglobal patterns from the forwarded high-frequent rules.Earlier we had proposed a model for synthesizinghigh-frequent rules on the basis of transaction population ofthe sites, support and confidence of the rule in the respectivesites. The rules that are forwarded by the local sites are“strong” rules which satisfy the minimum support andconfidence thresholds at respective sites. It is desired that thesynthesized rules from such forwarded patterns must closelymatch with the mono-mining results, ie. the results that wouldbe obtained if all the databases are put together and mininghas been done. When the rule is present in the site but fails tosatisfy the minimum support threshold value, it is not allowedto take part in the rule synthesizing process. In such situationsthe correction factor “h” plays a vital role in inferring theglobal support and confidence values. A suitable choice ofcorrection factor ‘h’ enables the domain expert to reap thevalid synthesized result. In this paper, the impact of correctionfactor in obtaining synthesized results close to themono-mining results is brought out.

Keywords

• Rule Synthesizing
• Transaction Population
• Correction Factor
• Weighting Model