From Multi -View Clustering to Ensemble Clustering

Multi-view clustering and ensemble clustering are two main paradigms in machine learning that leverage a dataset to reach different perspectives. Although, they share some simi-larities in dataset management, they differ in their methodology and application. This paper conducts a study to demonstrate that exploiting different sources of information and choosing the right aggregation method is an asset to improve the quality of clustering compared to using classical methods. Different views were generated on two benchmark databases. Then, a clustering algorithm was applied separately on the different views. Subsequently, three well-known ensemble clustering approaches were studied, namely majority voting, meta-clustering and ensemble clustering based on a co-association matrix. Thus, an analysis emphasizes the role of the fusion method used to combine the obtained partitions and its effect on clustering during this fusion. It also highlights the advantages of multi-view clustering over classical classification. Moreover, it shows that the diversity of partitions from different views can have a great influence on the clustering performance.