Probabilistic event pattern discovery

Abstract

Detecting occurrences of complex events in an event stream requires designing queries that describe real-world situations. However, specifying complex event patterns is a challenging task that requires domain and system specific knowledge. Novel approaches are required that automatically identify patterns of potential interest in a heavy flow of events. We present and evaluate a probability-based approach for discovering frequent and infrequent sequences of events in an event stream. The approach was tested on a real-world dataset as well as on synthetically generated data with the task being the identification of the most frequent event patterns of a given length. The results were evaluated by measuring the values of Recall and Precision. Our experiments show that the approach can be applied to efficiently retrieve patterns based on their estimated frequencies.