Research issues in outlier detection for data streams
Shiblee Sadik, Le Gruenwald · ACM SIGKDD Explorations Newsletter · 2014
In applications, such as sensor networks and power usage monitoring, data are in the form of streams, each of which is an infinite sequence of data points with explicit or implicit timestamps and has special characteristics, such as transiency, uncertainty, dynamic data distribution, multidimensionality, and dynamic relationship. These characteristics introduce new research issues that make outlier detection for stream data more challenging than that for regular (non-stream) data. This paper discusses those research issues for applications where data come from a single stream as well as multiple streams.