Symmetrically Relative Accuracy as Rule Evaluation Measure

• 발행기관: 한국자료분석학회
• 저자명: Hee Chang Park
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.16 No.4, 1727~1734쪽, 전체 8쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2014.08.30

영문 초록

Rule evaluation is one of the most popular techniques and plays an important role in data mining. A lot of measures have been proposed in different fields that try to evaluate features of the rules obtained by different types of association and classification tasks. The most frequently used rule evaluation measures are rule accuracy, sensitivity, relative accuracy, and relative sensitivity. In this paper, we propose a symmetrically relative accuracy to compensate the shortcomings of rule accuracy, sensitivity, relative accuracy, and relative sensitivity. This measure is divided the sum of rule accuracy and sensitivity to the sum of relative accuracy and relative sensitivity. It is a symmetric measure whose value is not changed by the position of antecedent and consequent item. And then we investigated the conditions of interestingness measures by Piatetsky-Shapiro (1991), and compared some properties of rule accuracy, sensitivity, relative accuracy, relative sensitivity, and symmetrically relative accuracy. The results showed that the symmetrically relative accuracy was more meaningful rule evaluation measure because it was symmetric, and monotonically increased as co-occurrence frequency increased.

목차

1. Introduction

2. Symmetrically relative accuracy

3. Simulation study by examples

4. Conclusion

References

키워드

relative accuracy relative sensitivity rule accuracy sensitivity symmetrically relative accuracy