Real-Time Traffic Load Prediction in Cellular Networks Using Cutting-Edge Machine Learning Approaches

Exponential growth of cellular traffic by the usage of smartphones and streaming services presents challenges to the maintenance of QoS-critical network performance, hence requiring accurate and efficient traffic prediction. This work extends the literature in cellular traffic load prediction with the application of machine learning. This paper discusses the reduction of dimensionality and the efficiency of models. Unlike the existing system that applies LightGBM, SVM, Linear Regression, and DBSCAN for clustering, the proposed system combines Random Forest, XGBoost, and a Stacking Classifier for higher accuracy and improved computational capacity. Collected key features include user throughput, cell throughput, latency, and resource block utilization. Preprocessing is done by filling missing values, normalization, PCA for dimension reduction, and DBSCAN clustering of similar traffic patterns. Mean Squared Error analysis showed that Stacking Classifier results in X% reduction in the prediction error, and Y% reduction in computation time compared with existing models. Random Forest and XGBoost improved feature selection that captured complex traffic patterns, while PCA and clustering reduced the training time without losing accuracy. The proposed framework can scale dynamically to meet the demands of 5G and beyond, providing robust adaptability to changing traffic conditions. In addition, the potential integration with real-time systems and anomaly detection could further improve network reliability.