A method for missing data interpolation by SVR

In this paper, an approach for interpolating the missing data by support vector regression (SVR) machine is proposed. First, the samples where some features are missing are separated from the original samples. Then the remaining samples are trained by SVR, where the feature values corresponding to the missing features are treated as the labels. Finally, the obtained hyper-surface is used to predict the missing features. Experimental results show the considerable effectiveness of the proposed method.