Detecting current outliers: Continuous outlier detection over time-series data streams

Abstract

The development of sensor devices and ubiquitous computing have increased time-series data streams. With data streams, current data arrives continuously and must be monitored. This paper presents outlier detection over data streams by continuous monitoring. Outlier detection is an important data mining issue and discovers outliers, which have features that differ profoundly from other objects or values. Most existing outlier detection techniques, however, deal with static data, which is computationally expensive. Specifically, for outlier detection over data streams, real-time response is very important. Existing techniques for static data, however, are fraught with many meaningless processes over data streams, and the calculation cost is too high. This paper introduces a technique that provides effective outlier detection over data streams using differential processing, and confirms effectiveness. © 2008 Springer-Verlag Berlin Heidelberg.