<p>This study investigates the applicability of using artificial neural networks (ANN) to predict the variations in P-wave velocity (<i>V</i>_p) as function of pressure (<i>P</i>) changes in carbonate rocks. Predicting this <i>V</i>_p–<i>P</i> relationship is critical for time-lapse seismic interpretation and geomechanics-related applications. Traditional laboratory measurements to determine <i>V</i>_p–<i>P</i> relationship are time-consuming, while existing empirical regression models often overlook the influence of petrophysical properties on <i>V</i>_p–<i>P</i> trends and lack adequate prediction accuracy. A comprehensive dataset of 363 carbonate core samples (1624 data points in total for measured <i>V</i>_p at varying <i>P</i>), covering diverse geological settings (different regions) and microstructures, was compiled from both new laboratory experiments and published studies. The ANN model incorporated petrophysical parameters including initial velocity, porosity, bulk density, mineralogy, and permeability. Results for the entire combined dataset demonstrate that ANN outperforms regression, reducing the root-mean-square error (RMSE) by up to 35% (from regression RMSE of 158&#xa0;m/s) when using initial velocity alone as an input. Incorporating petrophysical properties into ANN improved prediction accuracy, with further error reduction reaching an RMSE of 48&#xa0;m/s. ANN models trained on individual datasets achieved the lowest errors, highlighting their robustness for region-specific applications, while leave-one-out tests confirmed predictive reliability for unseen datasets. Despite the complexity of the <i>V</i>_p–<i>P</i> relationships in carbonates, this study shows the effectiveness of using ANN model to address such a problem when incorporating petrophysical rock properties as inputs. The study offers a workflow for integrating ANN-based method with petrophysics, potentially reducing experimental requirements while improving subsurface characterization.</p>