基于微粒群优化的连续属性离散化算法

连续属性的离散化是粗糙集理论的主要问题之一，也是影响粗糙集理论实用性的瓶颈之一。由于没有最佳离散化形式的统一标准，大多离散化算法采用的启发式带有较强的主观性，也难以得到较满意的离散效果。该文提出了基于微粒群优化的连续属性离散化方法，将各属性的离散化划分点初始化为一群粒子，在保证决策表分类能力不变的情况下，通过粒子间的相互作用寻求理想的离散化划分点，使得决策表引入较少的冲突。实验结果验证了该方法的有效性。

Algorithm for Discretization of Continuous Attributes Based on Particle Swarm Optimization

The discretization of continuous attributes is one of the main problems in rough sets and is one of the bottlenecks affecting the practicability of rough sets. Many discretization algorithms have been used at present, but there is not the complete criterion of the best discretization, it is difficult to get more satisfactory result for most algorithms. This paper presents an algorithm for discretization based on particle swarm optimization. It looks upon the position of one kind demarcation points as a particle to search for its best position on the premise of keeping the primary partition capability in the discrete decision table, and there is little conflicdve data. The experimental results prove the validity of this method.