An Algorithm for Attribute Impact Factor Analysis in Clustering Based on One Way Analysis of Variance ⋆

In order to analyse differences between clusters on each attribute quantitatively, and the significance that attributes have effect on clusters, this paper proposed an algorithm of analysing clustering results based on analysis of variance. We defined a metric about impact on clusters generated by single attributes or correlative attributes, through comparing differences of inra-clusters and inter-clusters about every attribute, and the impact degree can be regarded as impact factor of attributes in clustering. Compared with the limited application of traditional statistical methods, such as analysis of variance, logistic regression etc, experiments on real-world dataset indicated the suitability of this algorithm, and it can simultaneously deal with numeric and classical attributes, as well as analyse single and correlation attributes effectively.