Many methods were developed to get a correct velocity estimate of the subsurface. Erroneous velocities can cause incorrect positioning of reflectors, which thus affects the imaging quality. Technology has been developed to have better velocity estimation. From diving wave, and ray path tomography to full wave inversion. However, most modeling techniques have shown limitations in constructing an accurate velocity for the near-surface. Surface waves are used today in a wide range of near-surface modeling. It has shown effectiveness in overcoming hidden layers and velocity inversions within the near-surface. A large portion of the area within central Oman is covered by dunes. The quality of the surface wave picking depends on seismic data quality. The dune noise weakens the ground-roll energy. Furthermore, the fundamental mode within dune areas is biased since several modes were mixed, which affects the surface wave modeling process. Unsupervised machine learning was implemented to segregate the dunes area from the non-dune to apply different preconditioning for both areas. Then the areas were combined to build a guide function representing the dispersion profile at each location to drive the surface-wave picking. The inverted velocity using surface wave tomography shows a better correlation with topography when machine learning is applied. As it was able to capture the velocity variations within the area with a high resolution.