Efficient Mining of Dissociation Rules

Association rule mining is one of the most popular data mining techniques. Significant work has been done to extend the basic association rule framework to allow for mining rules with negation. Negative association rules indicate the presence of negative correlation between items and can reveal valuable knowledge about examined dataset. Unfortunately, the sparsity of the input data significantly reduces practical usability of negative association rules, even if additional pruning of discovered rules is performed. In this paper we introduce the concept of dissociation rules. Dissociation rules present a significant simplification over sophisticated negative association rule framework, while keeping the set of returned patterns concise and actionable. A new formulation of the problem allows us to present an efficient algorithm for mining dissociation rules. Experiments conducted on synthetic datasets prove the effectiveness of the proposed solution.