基于邻域的迭代最优化聚类算法的设计与分析

Design and Analysis on Iterative Optimization Clustering Algorithm Based on Neighbourhood

迭代最优化算法是模式识别中一种重要方法.算法随机确定k个分类中心进行初始类划分,再通过逐步求精的方法进行合理分类.通过对迭代最优化算法的分析和研究,指出该算法存在样本选择的盲目性、易陷入局部极值、没有考虑样本的聚类趋势等缺点.文中根据样本的聚类趋势,结合邻域思想,设计了基于样本邻域概念的迭代最优化算法,并对算法的时间代价进行了定量分析.该算法总的时间代价为O(n),已应用于网络管理中的知识分类中,并取得了满意结果.

The iterative optimization algorithm is an important method in pattern recognition. The parameters k, the center of class that will be elementary classified in original phases, is defined by random method in this algorithm. It is stepwise optimized and can achieve favorable results in patterns classification. By the researching and analyzing, the iterative optimization algorithm has some serious defects, which are selected samples blindly, presented local extremum in iterative optimization and don＇ t pay attention to clustering tendency of samples. According to the conception of the clustering tendency and neighbourhood of patterns, the newly algorithm, Iterative Optimization Algorithm Based on neighbourhood of Samples, is designed in this paper. The time complexity of the newly algorithm, which is O（ n ） and n is a number of samples in sets, is calculated in detail. This algorithm is applied in the knowledge classify of network management and acquired a satisfying results.