Hierarchical attribute reduction algorithms for big data using MapReduce

Attribute reduction is one of the important research issues in rough set theory. Most existing attribute reduction algorithms are now faced with two challenging problems. On one hand, they have seldom taken granular computing into consideration. On the other hand, they still cannot deal with big data. To address these issues, the hierarchical encoded decision table is first defined. The relationships of hierarchical decision tables are then discussed under different levels of granularity. The parallel computations of the equivalence classes and the attribute significance are further designed for attribute reduction. Finally, hierarchical attribute reduction algorithms are proposed in data and task parallel using MapReduce. Experimental results demonstrate that the proposed algorithms can scale well and efficiently process big data.