Machine Learning-based Congestion Prediction in Long Term Evolution Networks
No Thumbnail Available
Date
2026-08
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
In Ethiopia, LTE networks are currently the most widely used technology for mobile broadband services. This is mainly because LTE infrastructure has already been deployed in many areas and supports a large number of mobile users. Although 5G services have started in some areas of Addis Ababa, most mobile data users still depend on LTE networks. The continuous growth of mobile data consumption has increased the load on LTE cells, particularly in locations where many users access the network at the same time. When traffic becomes high, LTE cells may experience lower throughput and higher delay, which can negatively influence the QoS experienced by users. These problems are commonly observed in highly populated areas such as Addis Ababa, where network traffic changes depending on user activities throughout the day. This paper develops a Random Forest-based model for predicting LTE throughput congestion using network performance measurements.. The model uses different LTE Key Performance Indicators (KPIs) related to radio signal quality, resource usage, and traffic conditions. In addition, feature importance analysis is applied to determine which
KPIs have a stronger influence on throughput degradation and congestion.
Random Forest regression was chosen because LTE performance depends on several KPIs at the same time. The algorithm can learn the relationship between these KPIs and throughput variation. The results show that the model can estimate LTE throughput conditions and identify important network parameters related to congestion. The performance of the developed model is measured using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The results show that the model can estimate LTE throughput conditions and identify important networkparameters related to congestion. This research is based on historical LTE KPI data collected from selected eNodeB cells in the Bole Bulbula area of Addis Ababa. Future research may improve the model by including real-time KPI measurements and data collected from additional LTE sites or other wireless networks.”
Description
Keywords
LTE, Throughput Prediction, Congestion Analysis, Random Forest Regression, Machine Learning, eNodeB, KPI Analysis, Feature Importance