Blind protocol identification using synthetic dataset: A case study on geographic protocols

Abstract

Network forensics faces major challenges, including increasingly sophisticated cyberattacks and the difficulty of obtaining labeled datasets for training AI-driven security tools. Blind Protocol Identification (BPI), essential for detecting covert data transfers, is particularly impacted by these data limitations. This paper introduces a novel and inherently scalable method for generating synthetic datasets tailored for BPI in network forensics. Our approach emphasizes feature engineering and a statistical-analytical model of feature distributions to address the scarcity and imbalance of labeled data. We demonstrate the effectiveness of this method through a case study on geographic protocols, where we train Random Forest models using only synthetic datasets and evaluate their performance on real-world traffic. This work presents a promising solution to the data challenges in BPI, enabling reliable protocol identification while maintaining data privacy and overcoming traditional data collection limitations.