In this paper, the problem of detecting adversarial methods on the basis of mimicry against
locality-based classifiers has been studied in detail.In particular, the locality-based
mimicry by noise generation slightly outperformed the padding-based method, but both approaches
strengthened the importance of acquiring a representative set of observations for building a more
robust adversarial model.An exhaustive revision of the state-of-the-art
has been conducted, from which locality-based mimicry by action pruning and noise generation
were presented as effective methods for thwarting conventional machine-learning-based masquerade
detection capabilities.