Title: Soft clustering-based models
Authors: Mika Sato-Ilic - University of Tsukuba (Japan) [presenting]
Abstract: Researchers in the area of clustering data deal with large amounts of complex data. Soft clustering has the merit of explaining this data using a smaller number of clusters which enables us to obtain stable solutions in the sense of robustness and reproducibility. Model-based clustering is a framework of clustering methods that assume a model to the data so an adjusted partition can be estimated. Although this approach can obtain a clear solution as the result of the partition based on mathematical theory, we cannot avoid the risk that the assumed model might not adjust to the latent classification structure of the data. Therefore, we propose a framework called clustering-based models which exploits obtained clustering result as a scale of the latent structure of the data and applies it to the observed data. Since the cluster-based scale is obtained from the original data, the scale can measure the original data and then the re-measured data can be applied to the model to obtain a more accurate result. Soft clustering is utilized to capture the cluster-based scale to deal with the large amount of complex data and several soft clustering-based models with applications will be introduced.