摘要： 针对传统支持向量聚类的低性能和高耗费问题，提出最小二乘支持向量聚类(LSSVC)模型，设计自适应参数化方案。模型中包括两步簇划分算法和快速训练算法。前者对支持向量和非支持向量分别进行划分，后者采用增量方式，每次增量对应聚类模型的双向学习过程。实验结果证明，LSSVC可有效提高同类算法的效率，具有良好聚类能力，当数据增量为工作集大小的10%时，算法可在时间耗费和聚类准确率之间取得良好的平衡。

关键词: 支持向量聚类, 最小二乘, 双向学习, 自适应参数化

Abstract: Aiming at the bottleneck of poor performance and expensive consumption of traditional Support Vector Clustering(SVC), this paper proposes Least-Square Support Vector Clustering(LSSVC) model, and designs self-adaptive parameterization strategies. The model includes a new cluster labeling algorithm and fast training approach. The cluster labeling algorithm clusters Support Vectors(SVs) and non-SVs respectively. The fast training approach is implemented in incremental learning process, and after each data’s increment, a double-way learning procedure is conducted to adjust clustering model. Experiments demonstrate the improvement of LSSVC over its counterparts in efficiency and its competitive performance. And when the size of incremental data is 10% of the working set, it can balance cost and clustering accuracy well.

Key words: Support Vector Clustering(SVC), least-square, double-way learning, self-adaptive parameterization

中图分类号: 

• TP311.13