Researches On Multi-objective Attribute Reduction

Rough Set,which is put forward by professor Pawlak of Poland,is an effective method to deal with imprecise,inconsistent and incomplete data.After 30 years of development,rough sets has been widely used in fields of data mining,machine learning and decision analysis,etc.For the purpose of dimension reduction,attribute reduction attempts to obtain the minimum attribute set description of the information system from which redundant attributes are removed exhaustively.Many scholars have done lots of research in depth,contributing to formation of the attribute reduction system which dominates by a variety of reduction objectives.With the deepening of the research work,the research level of the system in theory and application has been continuously improved,but there still exists some deficiencies as follows:(1)Lack of effective heuristic attribute reduction algorithm for generalized decision preservation.(2)Reduction standard of distribution reduction is too restrictive,resulting in not consistent with requirements of some practical applications.(3)Core attributes are frequently used as starting point of forward greedy attribute reduction algorithms,but existing algorithms is more time consuming.Both theory and algorithm of multi-objectives attribute reduction system get perfect to some extent by some research work in this paper.Specific work as follows:(1)The research background and significance of rough set are given and the research progress of multi-objectives attribute reduction system in theory and algorithm are summarized respectively,the basic concepts of rough set theory are introduced briefly.(2)The discerniblity matrix based reduction algorithm for generalized decision preservation is introduced,considering that the time and space complexity of the algorithm is relative higher,so the local similarity degree and global similarity degree are presented and two kinds of heuristic algorithms for generalized decision preservation are designed respectively by using the global similarity degree as heuristic factor,experimental results indicate the effectiveness and efficiency of proposed heuristic reduction algorithms.(3)The [?,?] decision-confidence ordered pairs set is presented and the definition of generalized distribution preservation attribute reduction is given,the judgment method,solving algorithm of generalized distribution preservation attribute reduction are provided and we also discussed the relationship of the generalized distribution preservation reduction with other reductions in depth,experimental results indicate the correctness of relevant conclusions.(4)Lots of repeated computations is pointed out as one of reasons for low efficiency of the existing core computing algorithms,a mathematical model that can reduce the repeated computations to the greatest extent is constructed and a generic fast algorithm for computing core attributes is designed by using proposed mathematical model,experimental results show the effectiveness and efficiency of the proposed algorithm.