Filtrado de trazas MDT de alta movilidad mediante aprendizaje supervisado

Abstract

In beyond 5G networks, geolocated radio information will play a fundamental role to drive self-management algorithms in a zero-touch paradigm. Minimization of Drive Test (MDT) functionality provides operators with geolocated network performance statistics and radio events. However, MDT traces contain important location errors due to energy saving modes, which requires filtering out wrong samples to guarantee an adequate performance of MDT-driven algorithms. In this context, supervised learning (SL) arises as a promising solution to automate the design of MDT filtering procedures compared to rule-based solutions. This work presents a SL-based method to filter MDT measurements in road scenarios by combining user mobility traces and land use maps in the absence of labeled real user mobility traces. Assessment is carried out over real MDT data collected in a live LTE network. Results show that the model performs better in measurements with positioning accuracy information.