Towards TabTransformer-based Operating System Fingerprinting: A Preliminary Approach using the Nmap Database

Abstract

This work explores TabTransformers for operating system fingerprinting, leveraging their ability to learn relationships within data to make classification based on network traffic features. We utilise the Nmap database as the training dataset and evaluate a TabTransformer model on a hold-out test set. The optimised model achieves an outstanding AUC-ROC score of 0.9919, demonstrating its effectiveness in operating system classification. Expanding the number of different classes is one of the intriguing avenues for future work.