关键词: 数据挖掘, 维数灾难, 遗传算法, 属性规约, 分形

Abstract: Dimensionality reduction is the powerful method to tackle the“Curse of Dimensionality”.Genetic algorithms based feature subset selection technique is superior to traditional feature selection method in the dimensionality reduction of the high dimensional data set.However，it can not be used in the field of unsupervised learning such as clustering which has no class label to use.FDR（Fractal Dimensionality Reduction） is the new unsupervised feature selection method.But，it is infeasible in practice in the high dimensional data set for its multiple scanning of the data set and high time consume.Accordingly，the authors propose the GABUFSS（Genetic Algorithm Based Unsupervised Feature Subset Selection） algorithm which combines the genetic algorithm and the fractal dimensionality reduction technique to tackle the unsupervised feature subset selection problem in the high dimensional data set.The experimental results using synthetic and real life data set show that GABUFSS algorithm achieves better performance than FDR algorithm in the high dimensional data set and can find identical subsets additionally.

Key words: data mining, curse of dimensionality, genetic algorithm, dimensionality reduction, fractal