Clustering is a set of methods that are able to group data objects into clusters. It is an essential data mining tool and can be used in several applications. In some cases, collected data are represented using multiple data tables (or views). These views represent different feature types or different information categories describing the data. Therefore, multi-view clustering is crucial to analyze such data. This thesis provides an extensive comparative study of multi-view clustering. It considers ten benchmark multi-view datasets, eight multi-view clustering algorithms, and two classical clustering algorithms. We provide a concise methodology that includes the different aspects of experimental studies. First, we propose a new categorization of multi-view clustering approaches according to the recent progress in the literature. Also, we suggest a new hyper-parameter selection approach for multi-view clustering algorithms based on the intra-cluster distance adapted to multi-view datasets. We showed that the proposed hyper-parameter selection approach works efficiently and improves the clustering performances. Also, this thesis offers two new initialization approaches for multi-view clustering, which can be an alternative to the random initialization. Finally, this thesis also compares the selected multi-view clustering algorithms in terms of clustering results and execution time.