Beamtracking is a critical functionality in modern millimeter wave (mmWave) 5G New Radio (NR) systems and is expected to become even more critical in future 6G systems operating in terahertz (THz) frequency band. To enable uninterrupted connectivity base stations (BS) need to invoke this procedure periodically. Due to the use of massive antenna arrays in 6G THz systems, the amount of resources consumed by beamtracking will be extremely large making the time interval between sweeping beam configurations a very critical parameter. One of the phenomena affecting the choice of this interval is a user equipment (UE) micromobility - quick displacements and rotations of UE in the hands of a user happening even when the latter is in a stationary position. In this paper, by utilizing machine learning (ML) algorithms, we propose a procedure for the detection of the beam center at the BS side for applications characterized by different types of micromobility. We demonstrate that one can safely differentiate between applications characterized by low as well as distinctively different micromobility speeds. All the considered classifieds including the tree, random forest, and neural network perform qualitatively similarly. For applications having fast and similar micromobility speeds such as VR and gaming, the classification accuracy stays at around 85 - 90%. However, this loss in accuracy does not affect the ultimate goal of the remote application detection algorithm - understanding how often the beam alignment procedure must be invoked at UE and BS.