Capstone · Cybersecurity ML
Hybrid ML Intrusion Detection
A real-time hybrid intrusion detection system that runs an ensemble of ML models over network traffic and maps every alert to MITRE ATT&CK.
The challenge.
Signature-based tools miss novel attacks, and SOC analysts drown in alerts with no context about what an attacker is actually doing. The job was a detector that catches unknown behaviour and tells the analyst what it means.
The approach.
A hybrid NIDS and HIDS that runs an Isolation Forest, Random Forest and Autoencoder ensemble over CICIDS2017 traffic, with a real-time Streamlit dashboard. Live packet capture runs through Scapy; every alert is mapped to its MITRE ATT&CK technique so the analyst sees intent, not just a flag.
Engineering decisions.
-
Ensemble of three complementary models: Isolation Forest for unsupervised anomalies, Random Forest for known classes, Autoencoder for reconstruction-error outliers.
-
Every detection is mapped to a MITRE ATT&CK technique, so an alert reads as attacker intent rather than a bare label.
-
Live packet capture and feature extraction with Scapy feeding the models in real time.
-
Containerised with Docker and shipped as an interactive Streamlit dashboard anyone can run.
The outcome.
Combines anomaly detection and classification across CICIDS2017 attack categories, with MITRE ATT&CK mappings to support alert investigation. The Hugging Face demo requires a password; source code is available on GitHub.