Data mining methods for outlier detection are usually based on non-parametric density estimates in various variations. Here we argue for the use of local intrinsic dimensionality as a measure of outlierness and demonstrate empirically that it is a meaningful alternative and complement to classic methods.
展开▼
