Association and Casual rule mining using associative Bayesian networks

Association rule search is one of the ever increasing areas of the intelligent data analysis and data mining. Unfortunately, traditional approaches of this area often are not capable to cope with new challenging problems emerging while dealing with new classes of modern applications. The latter require new viewpoint on methodology and technology of association analysis where “classical” ones fail. For efficient solution of emerging tasks of the association analysis, the paper proposes “non-classical” model of probabilistic space and its fragment called “sub-defined probabilistic space”. While using algebraic view, the probabilistic models used are defined in terms of normalized Boolean algebra and lattices. Such a probabilistic model made it possible to cope with several challenging association analysis tasks. Between them, the proposed algorithm is capable of search for rare but “strong” association rules, mining negative rules of any forms and mining cause consequence rules. All these tasks are solved within the same framework called associative (algebraic) Bayesian network. The basic algorithm is demonstrated by simple case study, although the algorithm and corresponding software developed for this purpose were validated on an application of real life scale.