xStream detects outliers in feature-evolving data streams, where the full feature-space is unknown a-priori and evolves over time.
xStream is accurate in all three settings: (i) static data, (ii) row-streams, and (iii) feature-evolving streams, as demonstrated over multiple datasets in each setting.
xStream scales to over 1,000 point updates per-second (Intel Xeon® at 2.1GHz), while consuming bounded space.
Emaad Manzoor, Hemank Lamba, Leman Akoglu. Outlier Dete'x'ion in Feature-Evolving Data Streams. In 24th ACM SIGKDD International Conference on Knowledge Discovery and Data mining (KDD). 2018.