A New Approach to Rule Interest Measures

Rule extraction is one of the main tasks in the Knowledge Discovery process. Our hypothesis is that the interestingness of rules is strongly related to statistical independence between facts. Human reasoning assumes by default statistical independence and when a rule breaks this assumption it is considered interesting. This fact has been noticed long ago and even proposed as a principle for rule-interest measures. However, to the best of our knowledge, nobody has correctly developed this principle. Until now, when facing the evaluation of a rule as ~4 ~/3, where ,4 and B are complex conditions over a database universal relation, the way to evaluate the independence breaking principle was evaluating the fact that A and B were conditionally independent by using a function as