cic-ids2017

Deep learning framework for network intrusion detection using CNN, RNN (LSTM), and Hybrid architectures on CIC-IDS2017.

Real-time passive AI/ML detection of cyber threats in unidirectional IP traffic: read-only, no decryption, explainable alerts and a hash-chained evidence ledger.

Explainable intrusion detection on CIC-IDS2017 (2.8M flows). XGBoost at 99.65% accuracy with SHAP explanations for every prediction, including worked false-positive and false-negative case studies. MSc dissertation.

Local-first network security toolkit in Python: a port scanner with CVE lookups (NVD) and printable HTML reports, a web trainer for intrusion-detection models on your own labelled flow data (leakage-free splits, macro-F1, signed model bundles), and a CLI that predicts with them. Tested on Windows

Unsupervised anomaly detection: learns normal behaviour from network flows (association rules) and system-call traces (n-grams)

AI-based network intrusion detection system using Random Forest, CIC-IDS2017, Flask, and a web dashboard.

Session-level leakage in the standard 70/30 split of CIC-IDS2017, and what survives under a purged temporal split. ICAIRC 2026 (IEEE).

Network Intrusion Detection and Anomaly Scoring using Machine Learning

Memory-augmented SOC alert-triage research prototype with reproducible CIC-IDS2017 validation, leakage-aware controls, and publication-ready figures.

PFA project (EMSI 4th year): real-time network intrusion detection on 3M+ CIC-IDS2017 records - stacking ensemble of Random Forest, XGBoost and CNN-BiLSTM with Kafka and Spark Streaming.

Benchmark of six ML intrusion detectors on UNSW-NB15 and CIC-IDS2017: clean data, honest splits, measured results

Explainable hybrid IDS with leakage-aware cross-dataset evaluation on UNSW-NB15, CIC-IDS2017, and ToN-IoT.

Explainable ML intrusion detection on CIC-IDS2017 under leakage-aware evaluation

Multi-class network intrusion detection on CIC-IDS2017 — 99.85% accuracy across 15 classes on 2.5M flows. Random Forest, XGBoost, KNN vs. three from-scratch Bayesian classifiers, with SFFS feature selection and multi-seed stability analysis.

Threat detection that survives a one-way tap: no return path, no lookups, no blocking. On 877,801 real flows a conventional detector falls to 0.690 F1; 60 seconds of per-host state restores 0.981. Calibrated abstention, tamper-evident evidence, air-gapped console. SIH 2026 · SIH26145 · NTRO

Deep Packet Inspection Engine with ML-powered Intrusion Detection using CIC-IDS2017 dataset

Detect network intrusions with deep learning using CNN, LSTM, and hybrid models on CIC-IDS2017 data.

Unsupervised anomaly detection on network flows: synthetic NetFlow generator, clustering, Isolation Forest / LOF / OC-SVM / GMM / autoencoder, and a CIC-IDS2017 pipeline

Federated vs centralised network intrusion detection on original CIC-IDS2017 and corrected LYCOS-IDS2017 (code and results)

Explainable network-intrusion pipeline and an honest test of an LSTM world model on CIC-IDS2017: purged blocked CV, SHAP, offline Streamlit demo. Negative result included.