一种针对名义尺度变量的优化聚类算法

An Optimization-based Clustering Algorithm for Nominal Scale Variants

经典的聚类分析技术如系统聚类法和K-means等主要是处理间隔尺度的变量,而对于名义尺度变量则不适合。文章借鉴认知心理学和优化学习的思想,对名义尺度变量的聚类问题进行了研究,定义了名义尺度变量的距离度量——翻转距离,在此基础上,提出了一种目标函数优化制导的聚类算法,并演示了对名义尺度变量进行聚类的过程。实验表明,我们的算法结果可以得到合理的解释。

Traditional clustering techniques such as systemic clustering and K-means clustering adapt to interval scale variants, but not suit nominal scale variants. In this article, we study the clustering problem of nominal scale variants and define a kind of reversal distance to measure nominal scale basing on cognition psychology and optimization learning. By reversal distance measure, we propose a new clustering algorithm for nominal scale variants aiming at optimization of a target function, and then we demonstrate the clustering process of our algorithm. Experiments show that the result of our algorithm can be interpreted reasonably.