Study Of Attribute Reduction Algorithm Based On Genetic & Particle Swarm Optimization Algorithm And Rough Sets

Rough sets theory which is a powerful mathematical tool to deal with imprecise, incomplete and uncertain data. Rough sets attributes reduction is to remove the redundant attributes and preserve classification accuracy unchanged of information system. How to find the minimal attributes reduction is an important issue in rough sets. But to search the minimal attributes reduction has been proved to be a NP-hard problem; therefore, it could be valued to improve the algorithm efficiency at the same time to find the minimum attributes reduction. So this thesis devoted to studying attributes reduction based on Genetic algorithm (GA), Particle swarm optimization (PSO) and rough sets.Firstly, this thesis reviews the basic theories of rough sets, GA and PSO. Secondly, this thesis analyzes the algorithms of attributes reduction based on discernibility matrix, significance and dependability. Thirdly, this thesis analyzes the characteristics of rough sets attributes reduction based on GA and PSO. The advantages of attributes reduction based on GA and rough sets are widely search range, but its flaw are slow convergence and difficult to find the global optimal solution. The advantages of attributes reduction based on PSO and rough sets are fast convergence, the flaws are algorithm instability and easy falling into local optimal solution. Finally, this thesis studied of attributes reduction based on GA, PSO and rough sets, then proposed a new algorithm of attributes reduction.There are three features of this algorithm. The first is using a dependability of attributes as the heuristic strategies to solve the attributes core, and using the attributes core to restrict the initialized population in order to enhance the ability of local search, reduce the time complexity and improve the accuracy of the results. The second is that PSO increased the GA operations, which are selection, crossover and mutation, to fully use of the effective information of PSO. In order to raise the convergence speed, expand the search range, and make these two algorithms which each take what it needs. The third is that the algorithm defined the fitness function by the dependability and regulated the function parameters dynamically, in order to ensure the result is the minimum attributes reduction.