Proposition of Modified Balance Cross Entropy in Association Rule Mining

• 발행기관: 한국자료분석학회
• 저자명: Hee Chang Park
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.19 No.4, 1733~1741쪽, 전체 9쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2017.08.31

영문 초록

Interestingness measures for rule evaluation play important roles in association rule mining. In this paper, we proposed the modified balance cross entropy as association threshold. We also compared this measure with the entropy based measures through a few examples. In all cases, the cross entropy and the balanced cross entropy showed a tendency to decrease until a certain value, but to increase after that value. Also, they had zero or positive values, and did not have the direction of association. On the other hand, the modified balance cross entropy monotonically increased as co-occurrence frequency or co-nonoccurrence frequency increased. When two kinds of mismatch frequencies increased, the modified balance cross entropy monotonically decreased. It always took negative, zero, or positive values, and had the direction of association. Therefore the modified balance cross entropy is a better association measure than the cross entropy and the balanced cross entropy.

목차

1. Introduction

2. Modified balance cross entropy

3. Numerical example

4. Conclusion

키워드

association rule balance cross entropy cross entropy modified balance cross entropy data mining