Good and Bad Neighborhood Approximations for Outlier Detection Ensembles

摘要: Outlier detection methods have used approximate neighborhoods in filter-refinement approaches. ensembles artificially obfuscated to achieve diverse ensemble members. Here we argue that outlier models could be based on the first place, thus gaining both efficiency and effectiveness. It depends, however, type of approximation, as only some seem beneficial for task detection, while no (large) benefit can seen others. In particular, space-filling curves are approximations, they a stronger tendency underestimate density sparse regions than dense regions. comparison, LSH NN-Descent do not such construction ensembles.