An association rules algorithm based on kendall-τ
ICIC'11 Proceedings of the 7th international conference on Intelligent Computing: bio-inspired computing and applications
Hi-index | 0.00 |
Associative classification has arrested attention in recent years and made significant improvement in related applications. This paper introduces the concept of a new interestingness measure and examines its utility in some application domains. Many interesting-ness measures have been presented before with different qualities, which make them useful for some applications. Some of these measures, such as support and Interest, do not concentrate on all properties of an association rule. Besides, some of them, such as J_Measure and Mutual Information, have complex computes. We present a new geometric measure which uses all basic terms of a contingency table values P(A,B), P(A',B), P(A,B') and P(A',B') to estimate the association of itemsets A and B. The fundamentals of this measure are based on a simple fact: Since sum of these terms is constant, increasing each term causes the decrement of the other terms. Then, for better understanding, we describe our new measure in semi Cartesian coordinates. Finally, we demonstrate the benefits of using the new measure for association rule mining based on results obtained from a random generated dataset.