Extending an anomaly detection benchmark with auto-encoders, isolation forests, and rbms

Abstract

In this paper, the recently published benchmark of Goldstein and Uchida [3] for unsupervised anomaly detection is extended with three anomaly detection techniques: Sparse Auto-Encoders, Isolation Forests, and Restricted Boltzmann Machines. The underlying mechanisms of these algorithms differ substantially from the more traditional anomaly detection algorithms, currently present in the benchmark. Results show that in three of the ten data sets, the new algorithms surpass the present collection of 19 algorithms. Moreover, a relation is noted between the nature of the outliers in a data set and the performance of specific (clus-ters of ) anomaly detection algorithms.