AI based Network Attack Forecasting from Network Traffic Data

Source summary

Build a fully open-source, offline cyber-defence prototype learning temporal network-state transitions rather than classifying isolated flows. Use both flow-level and packet-level features from public labelled datasets, forecast future states/attack-stage probabilities and explain contributing features. Include reproducible training scripts/weights, a PCAP/CSV demonstration interface and comparison with a logistic-regression baseline using precision, recall, F1 and false positives. Architecture examples are optional, but interpretability and no cloud dependencies are required; supply short architecture/demo/presentation artefacts.