A Stacking Ensemble Technique to Predict Speed and Distance in 4G and 5G Communication Datasets

Understanding the circumstances and behavior of mobile users can significantly enhance the intelligence of mobile networks. Distance and speed of users constitute crucial components of the user profile. For example, predicting the user's speed can allow these networks to perform better resource allocation and identify high user-density areas and handover trends. We leverage machine learning to predict the user's speed and distance. Based on user behaviors, we define various User Equipment (UE) speed (or distance) categories, each including a range of speed values. We evaluate both solo machine learning (ML) algorithms and a stacking ensemble ML algorithm. We observe that the stacking-based ensemble algorithm, consisting of a decision tree and random forest classifier, provides the best accuracy. We evaluate our technique on three datasets viz., 4G-Passive, 5G-Passive, and 5G-Coverage. For speed prediction on 4G passive, 5G passive, and 5G coverage datasets, the accuracy values are 66.2 %, 88.7%, and 93.4 %, respectively. For distance prediction on the 4G passive dataset, the accuracy is 90.9 %. Our work demonstrates that it is possible to accurately profile the range of speed and distance using real-world, contextual data collected from 4G and 5G network deployments.