Batch-Free Event Sequence Pattern Mining for Communication Stream Data with Instant and Persistent Events

Communication systems consist of many subsystems and components among which various stream data including control messages as well as payload messages are transferred. Some messages can be regarded as events which are identifiable occurrence that has significance for system. Those events can be categorized into instant events and persistent ones according to whether they has duration in which some state is kept continuously. Instant events are treated as having no duration, while persistent events have some duration. Most conventional event sequence mining techniques do not consider the persistent events in which they treat persistent events as instant ones. Once persistent events come into play, event sequence patterns need to take into account occurrence constraints which indicate which persistent events are active when some instant or persistent event occurrence is observed. This paper proposes an event sequence pattern mining method which identifies frequent event sequences in which each event may have its associated persistent events as its co-occurrence constraints. The proposed method uses a sliding window technique which advances one event occurence at a time to get exact support count in the mixed stream of instant events and persistent events. It is equipped with an efficient pattern generation technique using dynamic programming technique, and an effecient counting technique for counting the occurrences of specific patterns. It has been implemented and evaluated for the experimental studies for data sets.