基于属性重要度的数据补齐方法

Attribute significance based imputation method

为更好地处理不完备决策信息系统的缺失值,提出一种基于属性重要度的增量式数据补齐算法。基于同类数据分布的内聚性和异类数据的耦合性定义属性的重要度,优先填补重要属性;提出一种属性重要度定义;采取加权欧氏距离和余弦相似从距离和角度两方面同时衡量样本的相似度;为避免不同类别数据的相互干扰,将相似样本的搜索范围限定为同类别的完备样本集。实验采用7个UCI标准数据集比较该算法和其它算法,实验结果表明,该算法能有效提高补齐后的分类识别率。

To better deal with missing values in incomplete decision information system,an attribute significance based incremental imputation was put forward.The attribute importance was defined based on cohesion and heterogeneity of data distribution of same category to design the filling order.Weighted Euclidean distance and cosine similarity distance were simultaneously employed to measure the similarity between samples.To avoid mutual interference between different categories of data,search scope was limited to complete samples with same category.Seven UCI standard datasets were used to compare the proposed imputation and other methods.Results show that this algorithm can effectively improve the classification performance.