基于改进粒子群优化的无标记数据鲁棒聚类算法

已有的聚类算法大多仅考虑单一的目标,导致对某些形状的数据集性能较弱,对此提出一种基于改进粒子群优化的无标记数据鲁棒聚类算法。优化阶段:首先,采用多目标粒子群优化的经典形式生成聚类解集合；然后,使用K-means算法生成随机分布的初始化种群,并为其分配随机初始化的速度；最终,采用MaxiMin策略确定帕累托最优解。决策阶段：测量帕累托解集与理想解的距离,将距离最短的帕累托解作为最终聚类解。对比实验结果表明,本算法对不同形状的数据集均可获得较优的类簇数量,对目标问题的复杂度具有较好的鲁棒性。

Improved particle swarm optimization based robust clustering algorithm for unlabeled data

Concerned at the problem that the most existing clustering algorithms only consider single object and they show pool performance in some datasets with particular shapes, a improved particle swarm optimization(PSO) based robust clustering algorithm for unlabeled data is proposed to resolve that problem. In the optimization phase: firstly, the classical formation of multi-objective PSO is adopted to generate the clustering solution set; then, the K-means algorithm is adopted to generate the random distributed initial population, and the random initial velocity is assigned to each particle; lastly, the MaxiMin strategy is adopted to decide the Pareto optimality. In the decision phase: the distances between Pareto optimal solutions and ideal solution are measured and the shortest one is selected as the final clustering solution. Compared experimental results show that the proposed algorithm show better clustering performance to datasets with different shapes and is robust to the complexity of objective problems.