Mixture models for high-dimensional clustering with applications to tumour classification, network intrusion, and text classification (2008-2010)

Cluster analysis is primarily used for finding groups in data of unknown structure. Its key applications can now involve data of very high dimension but only a limited number of experimental units. The aim of the project is to develop a mixture model-based framework for the clustering of high-dimensional data that can also handle feature selection, the choice of the number of clusters and their validation, the detection of outliers, and the use of labelled data in a semisupervised context. Key applications in medicine and technology will be studied, aiming at improved clustering performance and understanding of the underlying process.