Multi-task learning modeling of attribute mutual association based on confidence and imputation of missing values

Incomplete datasets with missing values increase the difficulty of data analysis. In this paper, the incomplete data are modeled based on the Auto-Association Neural Network(AANN) and the missing values are imputation. In order to strengthen the correlation among the attributes of samples, we propose an Attribute Mutual Associative Multi-Task Learning Model Based on Confidence (AM-MTL). By optimizing the data transmission path of nodes in the output layer, we construct an MTL architecture in which the main imputation task is parallel to the secondary classification task, and use the missing rate of sample attributes as the initial sample confidence to weaken the influence of incomplete input samples on the network parameter adjustment. The missing values and confidence are used as variables for training, and they can be updated iteratively, thus continuously reduce the estimation error of missing values. Experiments demonstrate the effectiveness of the proposed method in missing value imputation.