Dealing with Incomplete Data in Clustering

Over the years, significant developments have taken place in the direction of clustering numeric, categorical or mixed data. A new challenge is to cluster data with missing attribute values. The early algorithms used Fuzzy c-means to partition data into fuzzy clusters and estimate the missing values through estimation algorithms. Recently, Hathaway and Bezdek have proposed four strategies for effective clustering of incomplete data: Whole Data Strategy (WDS), Partial Distance Strategy (PDS), Optimal Completion Strategy(OCS) and Nearest Prototype Strategy(NPS). This paper provides a brief survey discussing each of the four approaches and the recent developments in the direction of opting them for treating incomplete data.