Edge-Aware Zero-Touch Slice Reconfiguration for Cross-Layer Resource Optimization in 5G Networks

Modern 5G networks require intelligent, autonomous orchestration to sustain ultra-low latency services and dynamic workloads. This paper presents a fully automated, cross-layer orchestration framework that integrates machine learning (ML)-based edge load forecasting with coordinated slice reconfiguration across edge, core, and RAN domains. A multivariate Random Forest regression model, trained on synchronized edge and RAN metrics, predicts CPU overloads 60 seconds in advance, enabling proactive, closed-loop orchestration—covering slice migration, UPF/SMF reassignment, and PRB profile adjustment through 3GPP-compliant interfaces, all without human intervention. The framework is validated on a real private 5G standalone (SA) testbed under diverse workload and stress scenarios. Forecast-triggered actions complete in 4.78 seconds on average, with service disruption under 1 second, demonstrating both responsiveness and compatibility with existing 5G deployments. This work delivers a practical, standards-compliant demonstration of predictive, zero-touch orchestration. Results confirm the effectiveness of forecast-driven, cross-layer adaptation for preserving service quality, and the approach is designed for scalable application to multi-UE, mobile, and large-scale 5G environments.