An Exemplification of Decisions of Machine Learning Classifiers to Predict Handover in a 5G/4G/3G Cellular Communications Network

Mobile applications and high-end Internet-of-Things (loT) devices are progressively becoming reliant on high-data rates and high endurance content delivery, while the provided data rates on cellular communications connectivity links are intrinsically time-varying. Handover (HO) is a primary element of cellular communication networks that requires to be appropriately managed considering that HOs among base transceiver stations and between access modes as a result of user mobility pose difficulties in delivering a desired rich user Quality-of-Experience (QoE). That is because there are multiple impediments to quality-of-service (QoS) such as reduced data rate and interruptions of service. The ability to make accurate decisions in predicting the upcoming HOs, and consequently the anticipated supported data rate, will give applications valuable latitude to make perceptive adaptations to evade substantial degradation in QoE/QoS. In this paper, we provide a proof-of-concept, and investigation of the decision accuracy of Machine Learning (ML) based prediction of mobile HOs in real-time in co-existing 3G/UMTS, 4 GIL TE, and 5G (NSA) networks. We render the results of our measurement campaigns, and relevant feature values capturing the influences of the environment in the form of GPS location coordinates, received signal strength indicator (RSSI), and associated connectivity modes. To that end, we conduct a countertype analysis to compare the performance of a distinct collection of ML classification algorithms to assess their decision accuracy in predicting HOs. Evaluation metrics revealed an optimum ML algorithm for HO prediction tasks, which should be useful to employ in the context of heterogeneous cellular communications networks.