基于AHTPSO的连续属性离散化算法

针对粗糙集不能较好地处理连续型属性的问题,结合粗糙集理论和粒子群算法,提出基于自适应混合禁忌搜索粒子群的连续属性离散化算法。首先,该算法通过对参数的自适应更新操作,从而避免了粒子群出现早熟的现象;然后将粒子群当代得到的全局最优粒子送入禁忌算法中进行优化,有效地提升了算法的局部探索能力;在兼顾决策表系统一致性的同时,将划分的断点初始化为一群随机粒子,通过改进后粒子群的自我迭代得到最佳的离散化划分点。实验结果表明,与其他结合粗糙集的离散化算法相比,该算法具有更高的规则分类精度和较少的离散化断点个数,对连续属性的离散化效果较好。

A discretization algorithm of continuous attributes based on adaptive hybrid tabu search particle swarm optimization

Aiming at the problem that the rough set cannot deal with the continuous attributes directly, combining with the rough set theory and the particle swarm algorithm, we propose a discretization algorithm based on adaptive hybrid tabu search particle swarm optimization. Firstly, the adaptive adjustment strategy is introduced, which cannot only overcome the problem of local extremum of the particle swarm, but also improve the abilities of seeking a global excellent result. In order to get the best global optimal particle, the tabu search is performed on the global optimal particle of each generation to enhance the local search ability of the particle swarm. Finally, under the premise of keeping the classification ability of the decision table, the attribute discretization points are initialized as a group of random particles. The algorithm searches for the best discretization points in the self iteration of the modified particle swarm. Experimental results show that compared with other discretization algorithms based on the rough set, the proposed algorithm not only has better classification accuracy and less discretization breakpoints, but also improves the discretization of continuous attributes.