Multi-objective selection approach for association mining based on interesting measures

摘要

In the era of digital information, the size of data collection has been growing significantly. Knowledge results in term of association rules obtained from the set of data are numerous and hard to select. This paper proposes the approach for selecting the interesting subsets of association rules from big association results. The selective criterion is based on well-known interesting measures including confidence, support and lift. The interesting measures are simultaneously considered in multi-objective context. While, confidence guarantees the accuracy of the association results. Support promotes popular association patterns and lift indicates rare association patterns. Thai stock market data in period of April 10, 2013 to September 5, 2014 were investigated and applied to the selection approach. The results showed that multi-objective selection approach reduces 246,084 association rules into 11 nondominated association rules.