Geo-locating UEs Using Multi-output Decision Tree Regressor
Dawood, Edel Goreil
MetadataShow full item record
Mobile communication networks constantly need to monitor and improve the quality of service and network coverage to their users, and to do so, radio measurements are required. One common method for gathering these measurements is Drive Tests (DT) in which vehicles equipped with measuring devices are sent to locations of interests. However, DT has several limitations one of which is that DT cannot be performed on indoor locations. Other approaches were later used; however, failed to prove a considerably higher efficiency. To address this need, mobile operators developed a standardized technique called Minimization of Drive Test (MDT) that take advantage of network users equipment (UE) to collect measurement such as geographical locations. While MDT is the de facto for collecting UEs measurement, it faces challenges affecting its usability and feasibility. To this end, in this thesis, we design and implement a novel system that can accurately predicts geographical information of UEs using machine learning regression techniques benefiting from MDT data that has already been collected. The proposed system achieves a state-of-the-art performance on predicting longitude and latitude coordinates with a mean square of error MSE of 0.0007 and 0.0013 respectively. While we have based our work mainly on the points with 5 RSRPS values, we have also created models that deal with points with 2,3, and 4 RSRP values.