面向聚类分析的自适应弹性网络算法研究

New adaptive elastic net method for cluster analysis

弹性网络算法是一种启发式算法,最初被提出是用来解决TSP(Traveling Salesman Problem)问题的,现如今,被广泛应用于聚类问题中,尤其对于高维空间数据聚类方面,有很大的优势。提出了一种新的自适应弹性网络算法(Adaptive Elastic Net,AEN)解决聚类问题,该算法利用弹性网络算法得到的K个中心点作为聚类初始中心点,并利用局部搜索择优算法在每次迭代中更新中心点。以聚类完成后每一簇的中心点到该簇元素的距离之和作为聚类质量评价标准,分别对随机生成的不同维度的50,100,300,500,1 000个数据点的数据集和UCI中多个标准数据集进行聚类,并将结果与传统聚类算法的聚类结果进行比较。实验表明:相较于传统的聚类算法,该算法可以有效地提高聚类质量。

The elastic net algorithm as a heuristic method initially applied to solve the traveling salesman problem,can be used as a tool for data clustering in n-dimensional space.In this paper,a new Elastic Net Approach(AEN)is proposed to solve the cluster analysis problem.The algorithm gets K points as the initial centroids and updates the centroids according to the local search preferred algorithm in each iteration.Simulation results are provided for instances of up to1,000points randomly generated in different dimensions,illustrating that the proposed method performs better than the traditional clustering algorithms in solution quality.