摘要
聚类分析本质是一个复杂的优化问题.为了准确地对真实数据集进行聚类,提出一种改进的多目标进聚类算法.该算法以多目标进化算法为框架,用谱聚类算法对样本特征进行转换,再将模糊C-均值聚类算法以及k均值聚类算法进行融合,以初始聚类中心作为迭代的自变量;在计算个体适应度部分加入对FCM、k均值聚类结果的重排序操作和排序结果融合操作,以保证聚类结果的多样性以及各类别之间的均匀性;此外,利用当代的最优个体以及历史最优个体分别与当代其他个体进行二次交叉,形成下一代的新个体,保证了群体的进化趋势;用爬山算法对多目标进化算法进行改进以快速找到最优解.在UCI数据集和人工数据集上的实验表明,该算法具有较高的准确性.
In order to accurately cluster real data sets,an improved multi-objective evolutionary clustering algorithm was proposed.Multi-objective evolutionary was used as the framework.First,the spectral clustering algorithm was used to transform the sample features,and the fuzzy C-means clustering algorithm and the k-means clustering algorithm were fused in this paper,the initial clustering center was regard as the independent variable of the iteration.Added the reordering operation and the fusion of the ranking results in the calculation of individual fitness to ensure the diversity of the clustering results and the uniformity among the categories.In addition,used the best individuals of the contemporary and the history to form new generations,which ensure the evolutionary trend of the group.And used the hill-climbing algorithm to improve the multi-objective evolution algorithm to quickly find optimal solution.Experiments on UCI dataset and artificial dataset showed that the algorithm has high accuracy.
作者
汪宏海
WANG Hong-hai(Tourism College of Zhejiang,Hangzhou 311231,China)
出处
《哈尔滨商业大学学报（自然科学版）》
CAS
2020年第5期570-575,共6页
Journal of Harbin University of Commerce：Natural Sciences Edition
基金
浙江旅游职业学院教改项目(基于ICVE平台小规模私有在线课程探索与实践—以《计算机网络管理》为例,项目编号:2020YB01)。
