Baselines. We choose seven state-of-the-art competitors in three categories to
show the outstanding performance of our proposed FDD and LAD. For kNN-based
algorithms, we choose Local Outlier Detection (LOF) [14] and Local Correlation
Integral (LOCI) [113]. Specially, LOCI provides an automatic, data-dictated cutoff
to determine whether an instance is an anomaly based on probabilistic reasoning.
For attribute-based methods, we include IForest [96] and Mass [138]. For
manifold-based methods, we choose two different manifold-based techniques used
in [2] including locally linear embeddings (LLE), and isometric feature mapping
(ISM), followed by LOF to obtain anomalousness measurement. We also include Strangeness
based Outlier Detection algorithm (StrOUD) presented in [9]. StrOUD
is based on Transductive Confidence Machines, which have been previously proposed
as a mechanism to provide individual confidence measures on classification
decisions [9].