In the last years, location-aware services and network management have driven the demand for user location estimation in mobile networks. Nevertheless, the location obtained from user terminals is not usually accessible to mobile operators. In addition, available cell Key Performance Indicators (KPI) vary highly from network to network, and only a few of them are always enabled widely. Currently prevalent Machine Learning (ML) based solutions have achieved high precisions, but they are bounded to a trained scenario, restricting their application to new areas. We propose a method for creating scenario-agnostic prediction models which solves the above problems by applying feature engineering, over a small set of easily obtainable KPIs, applicable for any ML method. Finally, the performance of the proposed method is demonstrated using a real network dataset.