Publication

Abstract : Future 6G Internet-of-Vehicles (IoV) systems must support three traffic classes: eXtreme ultra-reliable low-latency communication (xURLLC) for safety beacons, eXtreme enhanced mobile broadband (xeMBB) for infotainment, and eXtreme massive machine-type communication (xmMTC) for telemetry. IoV traffic is bursty and incident-driven, making fixed slice splits and mean-only forecasts unreliable and prone to service-level agreement (SLA) risk. This work proposes a risk- and fairness-aware slicing framework that converts traffic-forecast uncertainty into robust slice allocations. Per-slice demand is predicted using a LightGBM model that outputs point estimates and 95% prediction intervals through time-series cross-validation. These predictive distributions are transformed into safety-aware plans using Value-at-Risk (VaR), Conditional VaR (CVaR), and an optional DRO-based inflation term to guard against distribution shifts. A digital twin optimizer (DTO) maps these plans to xURLLC/xeMBB/xmMTC shares while enforcing slice floors and ceilings. It incorporates the Jain fairness index and uses EMA smoothing to stabilize allocations. A tunable risk parameter α and incident presets (accident, rain, parade) further adapt allocations by prioritizing xURLLC under high-risk conditions. Experiments on real IoV traces compare static, proportional fair (PF), forecast-only, and risk-aware DTO schemes. The proposed method reduces plan shortfall from 64% to 3%, maintains a Jain fairness score of 0.877, removes SLA violations, and yields negligible xURLLC tail-latency probability. Forecast accuracy reaches 95.91%, with well-calibrated uncertainty bands. The full pipeline is deployed in a Streamlit dashboard for operator use.