This paper presents a framework to identify and characterise anomalies in past en-route Mode S trajectories. The technique builds upon two previous contributions introduced in 2018: it combines a trajectory-clustering method to obtain the main flows in an airspace with autoencoding artificial neural networks to perform anomaly detection in flown trajectories. The combination of these two well-known Machine Learning techniques (ML) provides a useful reading grid associating cluster analysis with quantified level of abnormality. The methodology is applied to a sector of the French Bordeaux Area Control Center (ACC) during its 385 hours of operation over seven months of ADS-B traffic. The results provide a good taxonomy of deconfliction measures and weather-related ATC actions. The application of this work is manyfold, ranging from safety studies estimating risks of midair collision, to complexity and workload assessments of traffic when a sector is operated, or to the constitution of a database of ATC actions ensuring aircraft separation. This database could be used to train further ML techniques aimed at improving the state of the art of deconfliction algorithms.