Machine Learning SinFP Model Updates for OS Fingerprinting

Summary

Updates have been released for the Tenable MLSinFP model, which predicts a host's OS based on SinFP fingerprints, by rebuilding it on a newer tech stack, incorporating new features, and using a larger dataset, resulting in improved accuracy of 67%.

Change

Before this update, plugin 132935 “OS Identification: SinFP with Machine Learning” was targeting operating systems commonly seen up to January 2021; consequently any newer OSs were not available as predictions. Additionally, the plugin solely relied on TCP header information for model features.

After this update, the plugin targets operating systems commonly seen up to May 2025. Additionally the training dataset is larger (was 700K records, now 1.8M) and more varied (was 6K distinct SinFP fingerprints, now 100K), the predicted OSs names are cleaner and more consistent, and model features other than TCP header information are relied on. Ultimately these changes resulted in the plugin's balanced accuracy increasing to 67% (was 54%).

Impact

Remote detection of operating systems based on the MLSinFP method will have a slightly higher confidence score. Assets whose operating system was determined based on this method might have a different detected operating system.