Incremental imputation approach based on attribute significance and two similarity measuring functions

The switching power supply is widely used in many applications, but due to its complexity, it is difficult to solve the fault diagnosis problems by the traditional fault diagnosis methods. And many algorithms based on data driven, offer a tool to deal with these problems without any preliminary or additional information. However, the incomplete data (or missing values) deteriorate their performance. In this paper, a novel method of data imputation is proposed, which is based on attribute significance, k-NN and the gray relational analysis. Firstly, the instances with missing values are sequenced by the attribute significance; Secondly, we manage to find it´s k-nearest according to the grey relations; Finally, the missing values of the instance are imputed by the corresponding attribute values of referential observation. In order to demonstrate prove its performance, 6 UCI datasets are selected to compare with some other methods. Experiments show the proposed method get a better performance.