Summary Detection of wave types in a seismic record is generally done through expert judgment and can sometimes be subjective and erroneous. For instance, in seismic data acquired in foothills areas, the shot-gather is highly complex to interpret due to the scattering noise produced by the irregular topography and the heterogeneities in the near surface. We present a machine learning method based on K-means clustering to automatically detect regions on a seismic record and delimit scattering noise based on seismic attributes. We apply the method to synthetic data modeled by the finite difference method and real data acquired in Colombia. The results show that the proposed method can automatically separate the wave types in the shot gather.