Data-Driven Intelligence for Real-Time Traffic Prediction and Resource Allocation in Heterogeneous 5G Networks

Abstract

The emergence of fifth-generation (5G) cellular networks represents a transformative advancement in communication technology, offering ultra-low latency, high bandwidth, and the ability to support massive device connectivity. As mobile applications, Internet of Things (IoT) devices, and dataintensive services such as autonomous driving and augmented reality continue to expand, 5G is expected to serve as the foundation of next-generation digital ecosystems. While previous generations of mobile networks improved speed and efficiency, they still face challenges related to spectrum utilization, energy efficiency, and dynamic resource management. Traditional approaches, including static resource allocation and rule-based scheduling, are inadequate in highly dynamic environments, often resulting in network congestion, reduced Quality of Service (QoS), and inefficient spectrum usage. To address these limitations, this study proposes a data-driven framework that integrates machine learning techniques with 5G network analytics. A deep autoencoder is utilized to extract compact and meaningful latent representations from high-dimensional network performance data, effectively reducing noise and redundancy. These latent features are then evaluated using multiple machine learning classifiers, including K-Nearest Neighbours (KNN), Categorical Boosting (CatBoost), and Extreme Gradient Boosting (XGBoost), for baseline comparison. The proposed model, termed Deep Latent-Forest (DLF), combines the autoencoder with a Random Forest classifier to enhance prediction capability. Experimental results demonstrate that the DLF model significantly improves accuracy, robustness, and scalability in detecting dropped connections, making it highly suitable for real-time, intelligent 5G network performance management.