属性选择是一种有效的数据预处理方法.为了移除多变量时间序列属性集中的冗余属性和噪声属性,选择出包含足够原始信息并能提高精度的属性子集,提出一种基于相关性密度的属性选择算法.该算法使用相关性矩阵表示原多变量时间序列,定义每个属性的局部密度来表示属性的代表性,定义属性的判别距离作为该属性与其他属性间的区分度.最后根据决策图的分布来筛选具有较大代表性和区分度的属性.使用SVM分类器对UCI数据库中的4种不同数据集进行实验,实验结果表明该算法相比已有算法在分类准确度和时间效率上均有一定的优越性.%Attribute selection is an effective data preprocessing method.Aiming at removing redundant or noisy attributes from the multivariate time series attribute set and selecting an attribute subset containing enough original information to improve accuracy,an attribute selection algorithm based on correlation density is proposed.The algorithm employed in the correlation matrix to represent the original multivariate time series,the local density of each attribute to show its representative ability,the distance discriminant between attributes as their discriminant degree.Moreover,attributes with larger representativeness and discriminant degree were filtered according to the distribution of the decision graph.Experiments with SVM classifier on four different datasets from the UCI repository were performed.The experimental results demonstrate the great improvement of the proposed algorithm in classification accuracy and time efficiency when compared with the existing algorithms.
展开▼
