Unsupervised network traffic classification uses HDBSCAN and K-Means

This article details a method for unsupervised classification of network traffic to distinguish between different consumer brands operating under a single Autonomous System Number (ASN). The approach utilizes network-layer signals such as DNS resolver IP, IP range, and session counts, which are often overlooked by standard IP-to-ISP databases. By employing feature engineering techniques like wildcard IP cleaning, prefix extraction, and log transformation of skewed numerical data, the authors prepare the data for clustering algorithms like HDBSCAN and K-Means to assign brand labels without pre-existing training data. AI

IMPACT Provides a novel unsupervised approach for network traffic analysis, potentially improving brand attribution and network management.

RANK_REASON The item describes a novel methodology for network traffic classification presented in a research paper. [lever_c_demoted from research: ic=1 ai=0.7]

Read on Towards AI →

• hdbscan
• k-means clustering

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →