中性游走驱动的进化子空间聚类算法

Evolutionary Subspace Clustering Algorithm Driven by Neutral Walks

子空间聚类是解决高维数据聚类问题的有效方法之一,其面临的主要挑战是如何高效率地搜索一系列子空间.算法Chameleoclust利用可演化的染色体结构来搜索子空间聚类,取得了较好的聚类效果.但由于其以自然选择作为进化的主要驱动力,无法对进化过程中最主要的突变"中性突变"进行评价,因此缺乏足够的启发信息来引导搜索,导致搜索效率不高,且极易陷入局部最优等问题.本文提出一种中性游走驱动的Chameleoclust算法(Chameleoclust NW),该算法主要特点是以中性理论的思想为基础,将中性突变视为进化的主角,以进化潜力为启发信息对染色体进一步评价,并利用中性游走对算法搜索过程进行引导.实验结果表明,与Chameleoclust相比,Chameleoclust NW具有更高的搜索效率和准确率.

Subspace clustering is one of the effective methods for clustering high dimensional data. The main challenge of subspace clustering is howto search a series of subspaces efficiently. The Chameleoclust algorithm searches subspace clusters with an evolvable chromosome structure,and obtains good clustering results. However,due to taking natural selection as the main driving force of evolution,Chameleoclust cannot evaluate the neutral mutation,which is the most important type of mutations in the course of evolution.Therefore,there is insufficient heuristic information to guide the search,resulting in many problems such as lower searching efficiency and easily trapping in local optimums. Based on the neutral theory,this paper presents a newalgorithm Chameleoclust NW,which takes neutral mutation as the protagonist of evolution,and defines evolutionary potential as the heuristic information to further evaluate the chromosomes,and adapts neutral walks to guide the search process. The experimental results showthat when compared with Chameleoclust,the Chameleoclust NW algorithm has better performance in efficiency and accuracy.