A characteristic of most datasets is that the number of data points is much lower than the number of dimensions (e.g., the number of movies rated by a user is much lower than the number of movies in a dataset). Dealing with high-dimensional and sparse data leads to problems in the classification process, known as curse of dimensionality. Previous researches presented approaches that produce group recommendations by clustering users in contexts where groups are not available. In the literature it is widely-known that clustering is one of the classification forms affected by the curse of dimensionality. In this paper we propose an approach to remove sparsity from a dataset before clustering users in group recommendation. This is done by using a Collaborative Filtering approach that predicts the missing data points. In such a way, it is possible to overcome the curse of dimensionality and produce better clusterings. Experimental results show that, by removing sparsity, the accuracy of the group recommendations strongly increases with respect to a system that works on sparse data.