Modern methods of data analysis are rarely used in archaeology. Meanwhile, it is archaeology that opens up impressive opportunities for various interdisciplinary studies at the junction of archaeology, chemistry, physics and mathematics. XRF analysis, which has long been used to determine the qualitative and quantitative composition of discovered archaeological artifacts, among other things, provides arrays of digital information that can be used by machine learning methods for more accurate clustering or classification of artifacts. This is especially true for artifacts that are presented in the form of fragments of ancient ceramic amphorae or glass vessels. Such fragments, as a rule, represent the mass of the fragments mixed among themselves. There is a need to divide them into groups and then restore them as a single artifact from the detected fragments of one group. This paper presents a comparative analysis of the application of different clustering methods to combine artifacts into groups with similar properties.