Towards Identification of Network Applications in Encrypted Traffic

TLS fingerprinting, JA4, encrypted traffic, application identification, machine learning

Network traffic monitoring for security threat detection and network performance management is challenging because most communications are protected by encryption. This paper addresses the problem of identifying applications associated with Transport Layer Security (TLS) network connections.
We evaluate three primary approaches to classifying TLS traffic: fingerprinting methods, SNI-based identification, and machine learning based classifiers. Each method has strengths and limitations: fingerprinting relies on a regularly updated database of known hashes, SNI is vulnerable to obfuscation or missing information, and an AI technique such as machine learning requires sufficient labelled training data. To support research in this area, we have also created a novel dataset of labelled TLS communications for popular desktop and mobile applications.
The comparison of these methods that we present highlights the challenges of identifying individual applications, as TLS properties are significantly shared across applications. The simpler task of identifying a collection of candidate applications still provides valuable insights for network monitoring and can be achieved with high accuracy by all methods considered. Finally, we suggest practical use cases and identify future research directions to further improve application identification methods.