Hybrid Anomaly Detection Architecture for Cellular Networks with Human Validation Loop

Abstract

This project presents a Human-in-the-Loop Anomaly Detection Architecture for Big Traffic Data of Cellular Networks, aimed at enhancing real-time network security through a hybrid approach combining machine learning and human expertise. The system leverages flow-based traffic analysis and XGBoost algorithms to detect anomalies and classify service types from live packet data captured using Scapy. Unlike fully automated systems, this architecture integrates a human validation loop, where flagged anomalies are reviewed by experts to reduce false positives and reinforce model reliability. The feedback mechanism not only refines detection accuracy but also enables the model to evolve over time. Validated anomalies are securely logged with associated metadata in a structured database, ensuring transparency, auditability, and future reference. A real-time monitoring dashboard provides administrators with comprehensive insights into network behavior, alerting them to potential threats and supporting timely, data- driven decision-making. The architecture is designed to operate efficiently in high-volume, dynamic environments, offering scalability, adaptability, and resilience against evolving threats. By combining automation with human judgment, the system addresses the limitations of traditional models and provides a practical solution for modern cellular network anomaly detection, making it suitable for deployment in operational telecom infrastructures where accuracy, speed, and accountability are critical.