More and more software applications use human motions to improve the information retention. Some virtual environments are especially built to support the learning of human motions. However, these kinds of applications and their pedagogical feedback are rarely made from the analysis of 3D captured motions. This can be explained by the heterogeneity, the complexity and the high-dimensional nature of such data. However, machine learning techniques could be used to overcome these issues. This paper presents a first step towards the improvement of the human learning process of a motion, thanks to the analysis of clusters representing user profiles. In the context of the Bottle Flip Challenge and using raw captured motions, descriptors based on speed and acceleration are extracted. The motions are then automatically analyzed, according to two different approaches: one with the ground truth, and one without constraints on the number of clusters. The results suggest that the data are separable using the computed descriptors.