Wireless networks are widely used for a wide range of applications, from best-effort object tracking solutions to robot control in smart factories. However, the performance of these networks is highly dependent on their configuration. Worse, the different links have heterogeneous characteristics, and a homogeneous configuration is often suboptimal. We believe that digital twins are an excellent tool for achieving autonomous networks that can automatically reconfigure themselves based on conditions. To this end, digital twins of networks must be able to incorporate this heterogeneity into their models and capture the impact of a configuration on the performance of a given radio link. We therefore propose here a link-oriented prediction model, able to predict the expected Packet Reception Rate for a given MAC configuration. Our experimental evaluation demonstrates the relevance of a data-driven prediction method to capture the links specificities.