Symmetrically Weighted Net Confidence for Generation of Meaningful Association Rules

• 발행기관: 한국자료분석학회
• 저자명: Hee Chang Park
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.16 No.3, 1141~1149쪽, 전체 9쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2014.06.30

영문 초록

Today, government, public institutions and companies began to use data mining techniques to discover valuable information and knowledge from big database. Data mining is the process of analyzing data from different perspectives, and summarizing it into useful information through a huge volume database. Association rule, one of the well-studied methods in data mining, finds the relationship among itemsets in a massive database. In finding meaningful association rules, several objective interestingness measures are used, which are support, confidence, net confidence measure, and symmetrically pure confidence. But these measures are not sufficient to generate only interesting information, and have some drawbacks that they can not determine the direction of the association, and are difficult to interpret operationally. In this paper, we proposed a symmetrically weighted net confidence as an association threshold, and investigate the conditions of association criteria. Also, we compared this measure with some association thresholds through a few experiments. The results showed that the symmetrically weighted net confidence monotonically increased as co-occurrence frequency increased, had positive or negative values, and is symmetric.

목차

1. Introduction

2. Symmetrically weighted net confidence

3. Simulation data analysis

4. Conclusion

References

키워드

Association rule confidence net confidence measure symmetrically pure confidence symmetrically weighted net confidence