Automated Attribute Weighting Fuzzy k-Centers Algorithm for Categorical Data Clustering

Cluster analysis plays an important role in exploring the correlations in data by dividing datasets into separate clusters so that similar objects are located in the same cluster. Moreover, fuzzy cluster analysis can reveal the mixtures of clusters in datasets containing multiple distributions. Certainly, the outcome of clustering methods is approximately determined by the similarity definition. Thus, the similarity measurement is exceedingly important to the formation of fuzzy clusters. In fact, the similarity between two objects is mostly calculated by the mean of differences across multiple dimensions. However, the dissimilarity in some dimensions has little or no effect on the fuzzy clustering outcome. In this study, we explore such impacts for fuzzy clustering of data with categorical attributes. Accordingly, the impact of each attribute on each fuzzy cluster is calculated using an optimizer, and the overlapping dissimilar values are then adjusted by the corresponding weights. We propose to apply this approach to the Fk-centers clustering algorithm, and the experimental results show that our proposed method can achieve higher fuzzy silhouette scores than other related works. These results demonstrate the applicability of deploying of the proposed method in real-world application.