Accurately predicting ground vibrations during vibrodriving remains a significant challenge due to the risks posed to nearby structures. In the present study a data-driven approach to modeling soil vibrations using dynamic mode decomposition (DMD), an equation-free framework, is introduced. However, the performance of DMD is susceptible to noise, which may compromise its accuracy. To address the effects of noise, two DMD modifications are evaluated and compared: (i) augmented DMD (Aug-DMD), which uses time-delayed coordinates, and (ii) forward-backward DMD (FB-DMD), a denoising strategy that combines forward and backward time evolution to reduce systematic eigenvalue bias caused by noise. Synthetic geophone data augmented with white noise is used to simulate noise during the measurement process and to evaluate the effectiveness of the algorithms. In summary, the present study identifies the conditions under which each method predominates and provides recommendations for their application in real-time field scenarios.