A Triangulation-Based Approach for 4G LTE Networks Performance Evaluation Using Machine Learning

Abstract

In today’s digital world, the widespread use of smart devices and
high-bandwidth applications places significant pressure on 4G LTE
networks, leading to performance issues such as increased latency,
reduced throughput, network congestion, and service degradation.
To address these challenges, this study introduces a triangulation
approach for evaluating and optimizing 4G LTE network
performance using machine learning. Triangulation, in this context,
means integrating both multiple data measurements (data-level
integration) and multiple analytical algorithms (model-level
integration) to enhance result reliability, validity, and
generalizability. Using a real-world dataset from a Nigerian
network provider containing 910 data points across 257 features, a
composite performance metric TCMD (Throughput, Channel
Quality Indicator, Max Users, and Download PRB Utilization) was
engineered to represent network performance. Fourteen machine
learning models were evaluated, with Support Vector Regressor
(SVR) and Gradient Boost Regressor (GBR) identified as top
performers. A hybrid model combining both SVR and GBR was
then developed and applied. The results demonstrated notable
improvements, with the hybrid model achieving strong predictive
performance with R² = 0.74, MSE = 0.187, and MAE = 0.332. The
model successfully resolved approximately 70% of moderate
performance issues (bottlenecks) and improved the overall network
quality. In conclusion, the hybrid ML-triangulation framework
provides a reliable foundation for future LTE/5G performance
optimization and predictive network management.