5G Slicing with Machine Learning: Dataset Design, Model Adjustment and Performance Metrics

The rapid expansion of 5G networks has opened up transformative opportunities in enabling diverse applications such as augmented reality, immersive simulations, and real-time collaborative tools. Central to maximizing these opportunities is the concept of network slicing, a technique that partitions a single 5G network into multiple slices tailored to meet specific service requirements. However, the dynamic nature of these applications and varying network demands poses significant challenges in effectively managing these slices. This paper focuses on integrating machine learning (ML) techniques to optimize network slicing for a variety of use cases. A comprehensive dataset design is proposed, capturing use cases, such as high-throughput content streaming and latency-sensitive real-time applications. Additionally, the paper emphasizes ML model fine-tuning and hyperparameter optimization to ensure efficient resource allocation and enhanced quality of service. Performance metrics include model accuracy, responsiveness, and scalability, critical for real-time and latency-sensitive scenarios. Gradient Boosting Trees (GBT) emerged as the best-performing classifier, consistently achieving superior accuracy and scalability among the nine classifiers tested. This study provides a framework for incorporating ML-driven network slicing, demonstrating its potential to optimize resource allocation and ensure high-quality performance for diverse 5G applications.