A Comparative Analysis of Machine Learning Models for Cellular Load Prediction - Insights from Real-World Data

Accurate network traffic prediction is essential for efficient resource allocation and congestion management in telecommunications. This study evaluates the performance of various Machine Learning (ML) techniques, such as Linear Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), for predicting total traffic volume in a Long-Term Evolution (LTE) network. A real-world dataset collected from an LTE site of a telecom service provider in Kandahar Province of Afghanistan was used for model training and evaluation. Performance was assessed using RMSE, MAPE, and R² metrics. The results demonstrate that Deep Learning (DL) models, particularly LSTM, outperform traditional ML methods by effectively capturing temporal dependencies in traffic patterns. In contrast, traditional models exhibited lower predictive accuracy with high error rates. The findings highlight the potential of Recurrent Neural Networks (RNNs) for time-series forecasting in network traffic prediction, providing valuable insights for selecting suitable predictive models in similar datasets