Recognition and interpretation of regularly (e.g. every weekday) and irregularly (e.g. arbitrary events such as accidents) appearing traffic patterns in a road network are considered one of the most crucial questions in mobility data analysis. Knowledge of regular and irregular traffic patterns is a requirement for reliable traffic prediction or traffic control. In this paper, we present a spatiotemporal unsupervised machine learning approach using selforganizing maps (SOMs) for detecting regular traffic patterns at arbitrary intersections in a nationwide road network. The approach applies SOMs to traffic states expressed as gradual level-of-service (LOS) values, which were derived from travel time measurements of probe vehicles. For the identification of regular patterns, they were temporally categorized by daytime (60 minutes slots) and the day of the week. The approach consists of two steps: First, an unsupervised learning approach clusters intersections with similar time-dependent gradual LOS values in order to identify similar traffic patterns. Second, a subsequent temporal analysis enables the interpretation of temporal regularities of the patterns. Based on a one-year probe vehicle dataset, we showed that the clustering reveals plausible regularities for different intersections such as interchanges, urban intersections or roundabouts that are still interpretable by humans. Furthermore, the approach can be easily adapted to identify patterns in other parts of a road network.