Rough set theory has been used effectively in the steps of the process of data mining and knowledge discovery. In this simplified attributes reduction by rough set theory was important problem in data mining in general and in shortened particular attributes. In fact the data are often diverse, rich but sometimes can be excessive or inadequate, which affects knowledge discovery from data. In this paper, the author use a generalized discemibility matrix rough set model tolerances to build algorithms attribute reduction in set-valued information systems and illustrate the results of the algorithm through experimental program.