Dynamic imputation for improved training of neural network with missing values

作者：

Highlights：

• The proposed method aims to improve training of neural network with missing values.

• At each training epoch, imputed values are newly obtained using dynamic imputer.

• Imputation stopping rule is used to reduce computational cost.

• Diversifying imputations makes neural network more robust to imputation inaccuracies.

摘要

•The proposed method aims to improve training of neural network with missing values.•At each training epoch, imputed values are newly obtained using dynamic imputer.•Imputation stopping rule is used to reduce computational cost.•Diversifying imputations makes neural network more robust to imputation inaccuracies.

论文关键词：Neural network,Missing value imputation,Dynamic imputation,Imputation uncertainty

论文评审过程：Received 27 July 2021, Revised 23 November 2021, Accepted 2 January 2022, Available online 19 January 2022, Version of Record 21 January 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.116508