A comparative evaluation of outlier detection algorithms: Experiments and analyses

作者：

Highlights：

• Experimental comparison and analysis of unsupervised outlier detection techniques.

• Based on ROC, precision-recall, computation time, memory usage and robustness.

• Extend a nonparametric Bayesian method to model numerical and categorical features.

• Experiments make use of novel industrial datasets and assess generalization abilities.

摘要

•Experimental comparison and analysis of unsupervised outlier detection techniques.•Based on ROC, precision-recall, computation time, memory usage and robustness.•Extend a nonparametric Bayesian method to model numerical and categorical features.•Experiments make use of novel industrial datasets and assess generalization abilities.

论文关键词：Outlier detection,Fraud detection,Novelty detection,Variational inference,Isolation forest

论文评审过程：Received 2 March 2017, Revised 18 September 2017, Accepted 24 September 2017, Available online 25 September 2017, Version of Record 4 October 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.09.037