The growing complexity and scale of Internet systems coupled with the improved capabilities of Machine Learning (ML) methods in recent years have motivated researchers and practitioners to increasingly rely on these methods for data-driven design and analysis of wide range of problems in network systems such as detecting network attacks, performing resource management, or improving quality of service (QoS).

In this survey, we review a large number of prominent research papers that apply ML methods to design, analysis or evaluation of network systems. To this end, we divide these studies into the following six groups based on the area of network systems that they target: 1) Domain name system, 2) Application identification, 3) QoS, 4) Cloud services, 5) Network security, and 6) Traffic prediction. Within each group, we examine the type of ML methods as well as input datasets that are used by individual studies, describe how they address various challenges, and summarize their key findings. In particular, we explore how domain knowledge in networking can inform different aspects of applying ML techniques such as problem formulation, feature engineering, feature selection, and deployment. We summarize representative common practices in the networking community along with a number of remaining challenges and research gaps.