Crowdsourcing-enhanced missing values imputation based on Bayesian network

Abstract

Due to development of the Internet, the size of data continue to be large and rough. During the process of data collection, different kinds of data problems occurred, among where incompleteness is one of the most serious problems to deal with. The existing methods for missing values imputation have mostly relied on using statistics and machine learning. These methods are known to be limited in efficiency and accuracy, which are caused by high dimensional calculation and low quality of initial data. In this paper, we propose a new method combining Bayesian network and crowdsourcing to deal with missing values together. We use Bayesian network to inference missing values to improve efficiency while use crowdsourcing to obtain additional information in need to improve accuracy. Experiments on real datasets show that our methods achieve better performance compared to other imputation methods.