INTERNATIONAL JOURNAL OF INNOVATIONS IN SCIENCE & TECHNOLOGY AI-DRIVEN NETWORK TRAFFIC MONITORING AND VISUALIZATION SYSTEM USING A HYBRID XGBOOST AND DEEP AUTOENCODER FRAMEWORK

Abstract

Enterprise, cloud and IoT networks are generating lots of traffic, which are hard to monitor. We believe that the proposed XGBoost-Supervised Anomaly Detection (SA) and Deep Autoencoder-Unsupervised (UAE) based AI network traffic monitoring and visualization system would be beneficial. The system processes live packet streams, NetFlow/IPFIX files or historical PCAP data, aggregates packets into bi-directional flows, fetches 42 statistical and temporal data features, and uses a two-stage AI analytics pipeline. Real-time alerting and fastAPI backend and WebSocket topology views. Test is accomplished using the CIC-IDS-2018 benchmark and simulated live packet streams.  The proposed XGBoost classifier gets 99.45% accuracy and 99.45% F1-score on 1.8 ms of the previously unseen anomalous behavior. The reported processing components end-to-end are in millisecond level latency and the dashboard fully supports the quick interpretation of the network events. The outcome indicate that employing supervised classification, unsupervised anomaly scoring and real-time visualization can create a useful monitoring framework for security operations.