Incomplete Data Imputation Based on Single Output Sub-network Array and Missing Value Variable Description

Incomplete data widely exists in the real world, and it is of great significance to reasonably impute the missing values of incomplete data. In this paper, we construct a single output sub-network array of incomplete data, which can mine the association relationships between each dimensional attribute of incomplete data and other attributes. At the same time, to address the problem of incomplete input of the model caused by the presence of missing values, a new training scheme of the model is designed that treats missing values as variables and allows them to participate in network training together with existing values, thereby maximizing the use of known information. During the training process, network parameters and missing values are updated collaboratively, and the parameters between single output sub-networks are adjusted to each other by missing value variables. Finally, the training of the model and the imputation of the missing values are completed simultaneously. The experiment results on several public datasets demonstrate the effectiveness of the proposed method.