Handling Missing Data through Artificial Neural Network

Abstract

Missing data is a difficult problem to solve when working with real-world datasets. It is vital to improve data quality by imputing missing values in order to obtain effective data learning. Deep learning has recently risen to prominence as the most effective sort of machine learning technique for uncovering hidden knowledge in massive datasets and making accurate predictions. We have used “back-propagation artificial neural network” to impute categorical missing data. The main goal is to see how well “neural network” compares to statistics and machine learning for resolving categorical missing data. The results of “back-propagation” are compared with multiple imputation and random forest. It consistently outperforms alternative methods both in  “training” and “test” data sets, showing that “neural network” is a suitable for reconstructing “missing values” in “multivariate analysis”.