This paper provides an overall comparison between one-class classification algorithms and unsupervised outlier detection methods extended to the one-class classification scenario. The proposed methods' performances have been compared in terms of ROC AUC and AdjustedPrec@n. The Model Selection has been performed for each method through a 10 fold cross-validation procedure on the 80% of the data available. The remaining 20% has been exploited for testing purposes. The experiments have been conducted on 31 real-world datasets from the UCI Machine Learning Repository. SVDD and kNNglobal turned out to be the best performing algorithms for one-class classification scenarios.

10.1109/DSAA.2016.8

L. Swersky, H. O. Marques, J. Sander, R. J. G. B. Campello and A. Zimek, "On the Evaluation of Outlier Detection and One-Class Classification Methods," 2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA), 2016, pp. 1-10.

Davide Ilardi

University of Genoa