A Systematic Literature Review of Machine Learning Approaches for In-Browser Cryptojacking Detection

This paper analyzes the state-of-the-art machine-learning techniques for detecting in-browser cryptojacking. Our study follows a systematic literature review encompassing three phases: planning, conduction, and synthesis. We provide a comprehensive overview of the approaches, highlighting gaps, areas for improvement, and unique features while categorizing each technique. We identified 14 studies that propose novel methodologies. Most of these studies employed hardware metrics analysis, such as CPU instructions or usage, followed by network traffic analysis, hybrid models, and static feature analysis. The most commonly used algorithms were neural networks, convolutional and recurrent neural networks being the most prevalent, followed by support vector machines. These studies showed improvements in detection compared to traditional blocklist approaches, although they still faced issues with not being updated often enough. We found that the availability of a public and scalable dataset would benefit all research in this area. Additionally, few studies in network traffic analysis address privacy concerns.