Data imputation via conditional generative adversarial network with fuzzy c mean membership based loss term

There are some missing values in the data when the data is acquired from the sensors or other equipments. This makes it difficult for performing the analysis based on the data. There are two major types of existing methods for performing the data imputation. They are the discriminative methods and the generative methods. However, these methods are incapable for dealing the data either with a high missing rate or with an unacceptable error. This paper proposes an effective method for performing the data imputation. In particular, the conditional generative adversarial network (CGAN) is used to predict the missing data. Here, the enhanced fuzzy c mean algorithm is employed for performing the clustering so that the information on the local samples is exploited in the algorithm. The computer numerical simulations are performed on several real world datasets. Since this CGAN exploits the class of the missing values of the data, it is shown that our proposed method achieves a higher imputation accuracy compared to state of the art methods.