Nowadays, recent advances in modern techniques have resulted in data collections that are huge in both size and dimension. Such a trend emerges new challenges for statistical learning procedures designed to extract key information from a large amount of data. This thesis particularly addresses current challenges of unsupervised learning in the sense of data clustering and presents procedures that follow new trends in data clustering. Identifying a group structure of any real-world data becomes nowadays problematic due to several aspects. Firstly, it is well known that standard clustering procedures, e.g. k-means, are usually efficient on clean data, i.e. data without outliers, but the performance of such methods is highly affected by the presence of outliers deviating from the true underlying group structure. Hence, there is a need for a clustering method which is more robust against outliers. Furthermore, in some application domains, e.g. media domain, outliers as observations of high interest commonly form groups of very small sizes. In this context, not only the identification of such observations but also their group structure is required. Secondly, data clustering gets more difficult in high-dimensional space where the standard dissimilarity measures commonly fail. In order to overcome such limitation, dimension reduction or variable selection techniques are usually employed during data clustering. Finally, most existing clustering method commonly assume either specific group characteristics, e.g. group sizes, or even required for the number of clusters. Such assumptions might, however, be difficult to fulfill in case of real-world data. Although the main goal of this thesis is data clustering, the introduced clustering procedures additionally aim at outlier detection. For this reason, a discussion of identifying outliers in the context of a simple group structure is elaborated as well. The development of all introduced procedures is motivated by real application scenarios and the advantages of the methods are demonstrated on real-world data examples.