Analytical models combining methodology with classification model example

Distributed computing is nowadays almost ubiquities. So is data mining - time and hardware resources consuming process of building analytical models of data. Authors propose methodology of combining local analytical models (build parallely in nodes of distributed computer system) into a global one without necessary to construct distributed version of data mining algorithm. Basic assumptions for proposed solution is (i) a complete horizontal data fragmentation and (ii) a model form understood for human being. All steps of combining methodology are presented with classification model example in form of a rule set. Authors define and consider problems with combining local classification modelspsila rules into one final set of global model rules encompassing conflicting rules, sub-rules, partial sub-rules and unclassified objects. Algorithms for different combining strategies are also presented as well as their tests results. Tests were conducted with data sets from UCI Machine Learning Repository.