A novel fuzzy clustering algorithm based on kernel method and particle swarm optimization

作者：Chen Donghui; Liu Zhijing^*; Wang Zonghu

来源：Journal of Convergence Information Technology, 2012, 7(3): 299-307.

DOI：10.4156/jcit.vol7.issue3.35

• C-Means clustering
• Feature space
• Fuzzy C means clustering
• Kernel function
• Particle swarm

Fuzzy C-means algorithm (Fuzzy C-means, FCM) is one of the most popular fuzzy clustering method because of its simplicity and effectiveness, but FCM is sensitive to the initial cluster centers and noises. Possibility C-means algorithm (Possibilistic C-means, PCM) can pay for its freedom to ignore noise points. However, PCM is also very sensitive to initializations, and it sometimes generates coincident clusters. In order to overcome the weakness, a hybrid C-means fuzzy clustering algorithm, which combines FCM and PCM, is presented by the introduction of Mercer Kernel method and Particle Swarm Optimization (PSO). Firstly using Kernel method to map the original data from input space into Hilbert characteristic feature space by kernel transformation in which we can perform clustering efficiently, and calculates the distance between data points with a kernel function. In the second, using PSO to optimize the encoded data points, and low the influence of initialized data points. The most important is that PSO can avoid the disadvantage of easily falling into local optimum. The experimental results on standard data sets and synthetic data sets show that the proposed algorithm has less computational complexity, higher clustering accuracy and better global convergence.

• 出版日期2012
• 单位西安电子科技大学