Summary Today there is a wide menu of available unsupervised attribute ranking methods that can be applied to seismic data. However, their results can vary drastically. To have an insight of the differences between these methods for the specific case of surface waves classification, we analyze the behavior of unsupervised ranking methods on seismic attributes computed from synthetic multicomponent shot gathers generated for different near-surface models. The near-surface models include variations on: weathering layer thickness, irregular surface topography and irregular bedrock geometry. We use 17 seismic attributes, 12 near-surface geometries (divided into 4 categories) and 8 unsupervised ranking methods. We design a three-steps study to analyze the behavior of the clustering: (1) Generation of seismic attributes from the synthetic seismic data acquired in the near-surface model, (2) Performing ranking by using eight ranking methods for each near-surface model data, and (3) Analyzing the variations of the ranking depending on the near-surface model and the ranking method. We conclude that the ranking of the attributes highly depends on the ranking method. However, the ranking analysis shows that there are five attributes relevant in the clustering independently of the near-surface model.