AI-Driven intrusion detection system (IDS) for network traffic

Abstract

This thesis presents a real-time network intrusion detection system that integrates live flow capture via CICFlowMeter with a hybrid CNN–LSTM model. We apply Variance Inflation Factor (VIF) analysis to reduce an initial 83-feature set to 33 uncorrelated predictors, improving stability without loss of accuracy. The resulting CNN-LSTM achieves 98.75 % detection accuracy and 0.9993 AUC in benchmark fl ows, and processes live HTTP, SSH, DoS, and slowloris traffic with millisecond latency. Our work demonstrates practical deployment of machine learning–based intrusion detection system (IDS) in real networks, contributing a streamlined feature selection method and an end-to-end Python application for continuous monitoring.