Synthesizing Global Association Rules from Different Data Sources Based on Desired Interestingness Metrics

Since business houses are generally global, the required data for their corporate decisions are spread over multiple branches at different regions. In such circumstances, local pattern analysis-based global pattern discovery has become an efficient strategy for mining their multiple data sources. The traditional support-confidence framework alone is not enough for assessing the interestingness of synthesized global association rules. In this context, numerous interestingness measures have been developed in the past to meet various situations. Depending on the requirement, local branches and the central head may choose desired interestingness metric for evaluating local frequent-itemsets and global association rules, respectively. In this paper, we present a generalized synthesis procedure for synthesizing global association rules, based on any interestingness metric, from the mined local patterns forwarded by multiple data sources. We have also shown that the synthesized metric values are quite close to the targeted mono-mining results. Examples and experimental studies establish the validity of our proposal.