基于邻域粒化的小生境微粒群混合数据约简

混合决策系统中同时包含了符号型属性和数值型属性，经典粗糙集处理数值型属性时需要进行离散化，这样会造成信息的丢失。基于邻域粒化的思想，提出了小生境微粒群约简方法，分析了邻域距离函数的选择和大小对分类精度和约简属性数量的影响。邻域粒化的方法可以直接处理数值型属性，微粒群全局优化的特性可以有效的求解全部约简，小生境技术的采用避免了微粒群算法的早熟收敛。选取UCI数据集进行了仿真实验，结果表明该方法可以快速有效地求解混合决策系统的约简，而不影响系统的分类精度。

Hybrid attributes reduction based on neighborhood granulation and niche PSO algorithm

Hybrid decision systems include character attributes and numerical attributes. The lost of information when discretize the numerical attributes by Pawlak rough set is introduced. A reduction algorithm based on the neighborhood rough set model and the niche particle swarm optimization (PSO) algorithm is proposed. The affection of neighborhood operator to the reduction and classification is discussed also. Numerical attributes can be dealt directly by neighborhood relations. The PSO algorithm is a global optimization algorithm and can get all reductions. The use of the niche technology can avoid the premature convergence of the PSO. Experimental results demonstrate the validity and feasibility of the proposed algorithm, in application to four University of California at Irvine (UCI) machine learning databases.