The evolution of mobile networks has made optimizing the RF environment increasingly complex. Consequently, standards bodies and the industry have made a concerted effort to automate labor-intensive processes, such as drive tests. In this study, we present a method for extrapolating data obtained from the radio environment of a mobile network in the presence of coverage anomalies (samples outside the expected coverage area) that could introduce inaccuracies in conventional modeling. We also consider the data limited, sparse, and not entirely available throughout the area to be studied. Coverage representation is achieved through a hybrid technique that combines data extrapolation through Gaussian processes with a spatial kernel and modeling through artificial neural networks. The final results of this method are highly accurate in representing cells when limited data is available in the presence of anomalies and outliers.