MICAR: nonlinear association rule mining based on maximal information coefficient

Association rule mining (ARM) is an important research issue in data mining and knowledge discovery. Existing ARM methods cannot discover nonlinear association rules, despite nonlinearity being common and significant in engineering practice. Besides, negative association rules are less researched, although they can effectively reflect widely existing negative associations in practical complex systems. Consequently, we propose MICAR, a nonlinear ARM method based on the maximal information coefficient (MIC). MICAR can extract nonlinear association rules in positive and negative forms from transactional or continuous databases. MICAR is realized in three steps: data preprocessing, candidate itemset mining and association rule generation. MIC is used to identify the type of association rules and find potential nonlinear correlations. MICAR can also control the redundancy in itemsets and association rules by restricting their quantity and forms. Experiments on authentic and simulation datasets show that MICAR can extract high-quality positive and negative association rules more effectively and efficiently than existing methods, especially has the unique ability to extract nonlinear association rules.

Automated summary

Association rule mining (ARM) is an important research topic in data mining and knowledge discovery. This paper proposes a nonlinear ARM method called MICAR based on the maximal information coefficient (MIC), which can effectively extract high-quality positive and negative association rules, especially nonlinear association rules.