Different approaches to fuzzy clustering of incomplete datasets

Partially missing datasets are a prevailing problem in data analysis. Since several reasons for missing attribute values can be distinguished, we suggest different approaches for dealing with this common problem. For datasets, in which feature values are missing completely at random, a variety of approaches has been proposed. In other situations, however, the fact that values are missing provides additional information for the classification of the dataset. Since the known approaches cannot exploit this information, we developed an extension of the Gath and Geva algorithm that can utilize it. We introduce a class-specific probability for missing values in order to appropriately assign incomplete data points to clusters. Benchmark datasets are used to demonstrate the capability of the presented approach.