Binary Information Press
Outlier detection in data stream poses great challenges due to the limited memory availability and real time detection requirement. An outlier detection algorithm in mixed data stream based on the sliding window is introduced by clustering the data stream incrementally and generating the statistical information on behalf of the data distribution. When detection requirement arrives the multi-granularity deviation factors of some statistical information are calculated and the statistical information with high multi-granularity deviation factor is taken as the abnormal statistical information. At the same time a method determining the radius value of multi-granularity deviation factor is presented.