Autoassociative neural network for missing data imputation: A case study via the styrene production process

Abstract

AbstractA neural network‐based model is proposed to estimate missing values of incomplete datasets to augment their size. An autoassociative neural network (AANN), for which the output vector is identical to the input vector, was built for a styrene production process dataset. The proposed model was used to investigate the ability of an AANN to estimate one to three missing variables, evaluating the impact of the size of the datasets used and the level of correlation of the missing values with other process variables. Results show that the proposed AANN model can predict the process data even when the number of records used is relatively small. Moreover, the AANN method is suitable for estimating missing variables with an accuracy that depends on the correlation coefficient of the missing values with other process variables, keeping acceptable estimation for weakly‐correlated variables. Moreover, the model was tested on noisy data, and it is shown that the model trained on noisy data can also predict missing values in an acceptable estimation range.