A Bayesian scoring technique for mining predictive and non-spurious rules
Rule mining is an important class of data mining methods for discovering interesting patterns in data. The success of a rule mining method heavily depends on the evaluation function that is used to assess the quality of the rules. In this work, we propose a new rule evaluation score - the Predictive and Non-Spurious Rules (PNSR) score. This score relies on Bayesian inference to evaluate the quality of the rules and considers the structure of the rules to ﬁlter out spurious rules. We present an eﬃcient algorithm for ﬁnding rules with high PNSR scores. The experiments demonstrate that our method is able to cover and explain the data with a much smaller rule set than existing methods.