Multiple Imputation of Missing Data in Marketing

Abstract

Observations containing missing values are handled during data preprocessing phase. Marketing researchers have been handling the missing values in data mainly using statistical methods. Machine learning methods are infrequently used to handle missing data in the marketing domain. A systematic evaluation of treating missing data in marketing is required to verify if the current practices are indeed the best practices. We evaluate mean imputation, multiple imputation, sequential regression tree imputation and sequential random forest imputation on twenty real-world marketing datasets. Our results establish that multiple imputation and sequential random forest imputation perform better than the other methods under consideration.