Machine Learning in Cybersecurity: Clustering for Threat Detection

Read the blog: Machine Learning Clustering for Threat Detection

Author Credits: Alvin Wen, Software Architect, and Craig Chamberlain, Director of Algorithmic Threat Detection


Many modern standards, practices, and frameworks, including the MITRE ATT&CK matrix, emphasize the importance of discerning the unusual from the malicious in modern event logs and detections, which often contain many shades of gray between the interesting and the confirmed true positive threat detection. 

The MITRE ATT&CK matrix makes extensive recommendations to “baseline” normal activity. It contains at least 154 references to baselining normal activity, or monitoring for anomalous activity, as shown below. This guidance is necessary in order to discern the few malicious events from the great mass of benign events typically present in system logs. But MITRE ATT&CK does not discuss practical methods of undertaking baselining or anomaly detection at scale.

So what are we to do? This is where we need to add machine learning tooling to our threat hunting and detection practices. There are a number of powerful tools we can bring to bear, not to replace conventional detection, rather to supplement and assist it.

Article Link: Machine Learning in Cybersecurity: Clustering for Threat Detection