Machine Learning-Based Traffic Prediction in 4G LTE Networks. Case Study of a Mobile Operator in Cameroon
June 7, 2025
Mobile subscribers are increasingly demanding the availability of
broadband services while the radio resources allowing them to be connected
are limited. Understanding mobile Internet consumption trends and
subscriber traffic demands is essential to enable the management of existing
radio resources. However, it can be difficult to understand and describe the
data usage patterns of mobile users because of the complexity of mobile
networks. In this study, we study and characterize the data usage patterns and
user behavior in mobile networks to perform traffic demand prediction. We
exploit a dataset collected via a mobile network measurement and billing
platform of the Historical Telecommunications Operator (HTO) network
called U2020/MAE. We elucidate different network factors and study how
they affect data usage patterns by taking mobile users of the HTO as a use
case. Then, we compare mobile users' data usage patterns, considering total
data consumption, network access, number of sessions created per user,
throughput, and user satisfaction level with the services. Finally, we propose
an application that employs a machine-learning model to predict traffic
demand using the HTO data.
The evolution of mobile networks from 2-5G coupled with a permanent need for broadband services by mobile subscribers has led to a significant surge in mobile internet resources and traffic. optimization of the available resources globally (Guo et al., 2018; Silva et al., 2018).The usage of applications like video streaming in both normal and high definition, online gaming, online conferences, and meetings by mobile subscribers has deeply improved this growing mobile data traffic. Many reports demonstrate that an important part of Internet traffic generated from mobile users' equipment is due to multimedia content (Gember et al., 2011; Huang et al., 2013; Maier et al., 2010; Shafiq et al., 2011). According to Ericsson (2020) report, video content alone accounted for 60% of mobile data traffic, with projections suggesting that this figure will rise to 74% by 2024 (Ericsson, 2018b). Additionally, it was anticipated in 2018 that by 2022, global traffic resulting from mobile data consumption will be twelve times higher than it was in 2018 (Ericsson, 2018a). Mobile carriers have evolved into a complicated entity designed to meet the evergrowing demand for mobile traffic (Damnjanovic et al., 2011). The rising demand for mobile data, coupled with increasing network complexity and the expanding number of connected users, presents significant challenges in both the control plane and user plane for radio resources management. Understanding mobile users' data usage patterns is a problem for content providers as a result of the increase in mobile users and mobile traffic demand. Mobile network operators then have the obligation to efficiently manage the available resources based on the data usage consumption and behavior of their subscribers. According to research and common usage of mobile devices, their energy consumption is greatly impacted by the type of active applications and their respective data usage pattern (Huang et al., 2013). It is crucial for service and content providers, as well as end users, to understand the data usage trends and behavior of mobile users across various markets and geographical areas. Mobile network carriers can use this information to predict the growing demand for mobile data usage (Cisco, 2017), and to do proper capacity planning and efficient network
After implementing each of the previously presented machine learning algorithms, all models achieved a certain level of accuracy, as detailed in Table (2). The LSTM-based model outperforms the others with an accuracy of 95.88%, making it the preferred choice for making predictions. The results of the comparative analysis are summarized in Table (3). We can conclude that the LSTM-based model offers superior performance and will be used for the development of the prediction tool.