Fuzzy c-means clustering of partially missing data sets based on statistical representation

The fuzzy c-means algorithm is a useful technique for clustering real s-dimensional data, but it can not be directly used for partially missing data sets. In this paper, the problem of missing data handling for fuzzy clustering is considered, and a statistical representation of missing attributes is proposed. The approach reduces the statistical analysis of missing attributes to the subsets of the dataset with similar data of incomplete data to impute the missing attributes, thus is helpful in enhancing the learning of missing attributes and the performance of fuzzy clustering based on the recovered data. Comparisons and analysis of the clustering results of the incomplete IRIS data demonstrate that the proposed statistical representation can estimate missing attributes rationally and improve the fuzzy c-means clustering of incomplete data.