Put in place by the International Maritime Organization, the Automatic Identification System is a worldwide maritime electronic system that sends radio broadcasted messages at a high rate between the stations, either on board the vessels or on shores. However, some misuses of the system such as identity theft, localization spoofing or disappearances have been demonstrated. The high rate of transmission implies a considerable amount of data to process in order to point out those irregularities. This paper proposes a method based on data mining and clustering methods combined to an integrity assessment of AIS messages for anomaly detection, with a proposition of software architecture for a data processing done both on-the-fly and with archived data. The computation of confidence coefficients and the use of data mining techniques will lead to behaviour characterization with the purpose of enhance the maritime situational awareness.