A Study of the Causal Strength Measures in a Viewpoint of Association Rule Threshold

• 발행기관: 한국자료분석학회
• 저자명: 박희창(Hee Chang Park)
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.18 No.3, 1199~1206쪽, 전체 8쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2016.06.30

영문 초록

By Wikipedia, data mining is the computational process of discovering patterns in a big data involving methods at the intersection of artificial intelligence, machine learning, statistics, and database systems. In this paper, we explored the features of some probabilistic measures of causal strength in a viewpoint of association rule threshold, and compared these measure with confidence to be the most frequently used. Furthermore, we checked the conditions of interestingness measures and some useful properties, and compared some properties of causal strength measures through a few experiments. All of causal strength measures increased as confidence increased. These measures monotonically increased as co-occurrence frequency increased. And also we discovered the fact that Eells, Suppes, and Galton measures had negative or positive values and, Lewis measure and Good measure had values to be less than or greater than 1. Thus probabilistic measures of causal strength are better than confidence because these measures have the direction of the association.

목차

1. Introduction

2. Probabilistic measures of causal strength

3. Numerical example

4. Conclusion

References

키워드

Association rule causality causal strength measure confidence data mining