In modern radio networks with large antenna arrays and precise beamforming techniques, accurate user positioning plays a key role in enabling seamless mobility management, link optimization, navigation and safety control. In open and rural areas, Global Navigation Satellite Systems (GNSS) are able to provide high-accuracy and high-reliability positioning performance. However, in urban and densely built-up areas the GNSS performance is typically substantially degraded due to rich scattering and multipath propagation effects. In this paper, we propose a machine learning based solution to boost positioning accuracy in urban areas by (i) obtaining User Equipment (UE) position from beamformed Radio Signal Strength (RSS) measurements and (ii) coherently fusing it with GNSS-based positioning data to enhance overall positioning performance. Based on the obtained numerical results, we were able to achieve a meter-level accuracy with the proposed machine learning model utilizing the beamformed RSS measurements, and subsequently improve the positioning accuracy further via fusion with GNSS data.