RAVA: A Human-in-the-Loop Hybrid Platform for Explainable Anomaly Detection and Diagnosis in LTE RANs

Today, Deep Learning (DL) has significantly improved anomaly detection accuracy in LTE Radio Access Networks (RANs). However, a gap remains between theoretical performance and real-world deployment. Traditional monitoring tools lack the interactivity and interpretability necessary for quick troubleshooting and resolution. In this paper, we present RAVA (Real-Time Anomaly Visualisation and Analysis), a comprehensive web platform that operationalises AI-driven detection. RAVA proposes a modular architecture that separates heavy inference tasks from visualisation, utilising WebSockets for real-time responsiveness. Our system includes a diagnostic engine that converts SHapley Additive exPlanations (SHAP)-based feature importance into actionable root causes, such as congestion and interference, through a deterministic, rule-based module. Additionally, a Human-in-the-Loop (HITL) feedback mechanism allows engineers to validate detections and actively improve the ground truth repository for ongoing system enhancement. The platform is tested with a real-world dataset of 1,650 eNodeBs. While relying on the underlying engine, which achieves a 93.89% F1-score, this study demonstrates its operational usefulness through a detailed case study of a site failure. Results indicate that RAVA effectively bridges the gap between black-box AI and engineering workflows. This platform reduces time-to-insight and supports closed-loop network management.