Wave peak particle velocity (PPV) is an index often used internationally to determine the structural damage caused by mining blasts. Since detailed spatial variations due to local soil effects are ignored in empirical PPV equations, new artificial intelligence (AI) methods have become a proper tool for describing the local effect of wave propagation in complex areas. An artificial neural network (ANN) has been developed in this work to overcome traditional empirical method limitations. Data collected from several seismic stations strategically located within a pilot study area, where two controlled explosions were designed and detonated exclusively for this research, were used to train an ANN model and a non-linear regression (NLR) model. The goodness of fit of the two methods applied in this research, as well as the spatial variation of the site effects, were analyzed. Results showed that the ANN model has a high capacity to adapt to the local soil effects, overcoming the problems related to regression traditional methods and offering new lights for understanding the soil and topographic amplification phenomenon.