基于Relief算法的特征学习聚类

Feature Learning Clustering Based on Relief Algorithm

聚类作为数据挖掘常用工具之一,是按照事物间的相似性进行的一种无监督分类.然而传统的聚类方法较少考虑特征权值.为此,通过研究、分析Relief算法及其在聚类应用中存在的问题,提出了一种基于Relief算法的特征评价函数,并将此函数运用到特征学习聚类中,以解决特征权值取值不当对聚类产生的负面影响.

Clustering unsupervised classification according to similarity of objects is one of common tools in data mining. Unfortunately, many traditional clustering algorithms assumed the distribution of each feature is uniformed. By researching and analyzing relief algorithms and its weakness in clustering, a novel feature criterion function based on relief algorithm has been proposed in the paper. Meanwhile, we apply the function into feature learning clustering in order to counteract the negative affects by the given feature weighting wrongly.