In today's diverse cyber threat landscape, anomaly-based intrusion detection systems that learn the normal behavior of a system and have the ability to detect previously unknown attacks are needed. However, the data gathered by the intrusion detection system is useless if we do not form reasonable data points for machine learning methods to work, based on the collected data sets. In this paper, we present a survey on data points used in previous research in the context of anomaly-based IDS research. We also introduce a novel categorization of the features used to form these data points.