Fuzzy belief pattern classification of incomplete data

Fuzzy c-means (FCM) algorithm is popularly used as a tool in various classification applications, yet it cannot be applied directly to a data set with missing values. Unfortunately, it is inevitable that a real world data set always contains missing values. Consequently, finding an effective way to handle incomplete data becomes an essential and important task. In this paper, an approach based on combining FCM and Dempster-Shafer theory is proposed. FCM is used as a preprocessing unit to obtain the initial degrees of belief on complete data and to construct pieces of evidence in a decision rule set. Then Dempster-Shafer theory is applied to make the final decision on which class incomplete data should belong to. It shows that the combined method achieves a better classification result compared with a popular imputation algorithm.