The Research On Rough Set Attribute Reduction Algorithm Based On Niche Discrete Differential Evolution

Efficient attribute reduction algorithm is the necessary foundation of applicationin the field of intelligent decision and data mining with rough set theory. Someresearchers have proven that it is hard to find the minimal reduction of aninformation system because of the explosive growth of a property combination, andthere is no unified and standard reduction algorithm. Therefore, to explore moreefficient attribute reduction algorithm to quickly find the optimal reduction orsuboptimal reduction is an important issue of rough set theory.This paper first elaborate on the rough set theory, differential evolution theory,niche theory and some general methods to decision table’s attribute reduction inrough set theory, such as attribute reduction algorithms based on discernibilitymatrix, attribute reduction algorithms based on discernibility function and attributereduction algorithms based on attribute dependency and so on, with special emphasison analysis of differential evolution reduction algorithm base on the attributedependency. At the same time, also analyze the advantages and disadvantages ofeach algorithm.Then, on the basis of analysis and comparison, to deal with the problems ofpremature convergence, a novel attribute reduction algorithm is proposed—the nichediscrete differential evolution reduction algorithm based on attribute dependency.The main feature of the algorithm is the introduction of niche that is the concept ofbiology to attribute reduction. On the one hand, the new algorithm uses nicheexcluded mechanism as like as eliminative method. On the other hand, in order tomaintain the diversity of the population, the new algorithm introduces a penaltyfunction to adjust the fitness of individuals in the population, so that the individualsevolve in different living environments and more relative minimum reductions canbe found.Finally, the analysis and comparison of experiments proved that the nichediscrete differential evolution reduction algorithm based on attribute dependency iseffective and has its own advantages on finding more relative minimum reductions.