Interval clustering algorithm for fast event detection in stream monitoring applications

• Clustering can be used for fast event detection over continuous data streams.
• We propose an interval clustering algorithm which observes a user-given error bound.
• The error control is necessary for critical applications such as patient monitoring.
• Our error estimation measure can be used for any types of input data.
• It can also be used as a distance measure in a clustering algorithm.

In stream monitoring applications, it is important to identify rapidly abnormal events over bursty data arrivals. By clustering similar conditions used in event detection, it is possible to reduce the number of comparisons and improve the event detection performance. On the other hand, event detection based on these clustered conditions can produce inaccurate results. Therefore, to use this method for critical applications, such as patient monitoring, the number of event detection errors needs to be kept to within a tolerable level. This paper presents an interval clustering algorithm that provides an error control mechanism. The proposed algorithm enables a user to specify a permissible error bound, and then uses the bound as a threshold condition for clustering. The simulation conducted based on real data showed that the algorithm improves the performance of event detection by clustering conditions while observing a user-specified error bound.