Explainable Distance-Based Outlier Detection in Data Streams

Explaining outliers is a topic that attracts lot of interest; however existing proposals focus on the identification relevant dimensions. We extend this rationale for unsupervised distance-based outlier detection, and through investigating subspaces, we propose novel labeling in manner intuitive user does not require any training at runtime. Moreover, our solution applicable to online settings complete prototype detecting explaining data streams using massive parallelism has been implemented. Our evaluated terms both quality labels derived performance.