Title: QoS-Aware Load Balancing in Wireless Networks using Clipped Double Q-Learning

Abstract: In recent years, long-term evolution (LTE) and 5G NR (5th Generation New Radio) technologies have showed great potential to utilize Machine Learning (ML) algorithms in optimizing their operations, both thanks to the availability of fine-grained data from the field, as well as the need arising from growing complexity of networks. The aforementioned complexity sparked mobile operators' attention as a way to reduce the capital expenditures (CAPEX) and the operational (OPEX) expenditures of their networks through network management automation (NMA). NMA falls under the umbrella of Self-Organizing Networks (SON) in which 3GPP has identified some challenges and opportunities in load balancing mechanisms for the Radio Access Networks (RANs). In the context of machine learning and load balancing, several studies have focused on maximizing the overall network throughput or the resource block utilization (RBU). In this paper, we propose a novel Clipped Double Q-Learning (CDQL)-based load balancing approach considering resource block utilization, latency and the Channel Quality Indicator (CQI). We compare our proposal with a traditional handover algorithm and a resource block utilization based handover mechanism. Simulation results reveal that our scheme is able to improve throughput, latency, jitter and packet loss ratio in comparison to the baseline algorithms.