How to introduce expert feedback in one-class support vector machines for anomaly detection?

摘要

Anomaly detection consists of detecting elements of a database that are different from the majority of normal data. The majority of anomaly detection algorithms considers unlabeled datasets. However, in some applications, labels associated with a subset of the database (coming for instance from expert feedback) are available providing useful information to design the anomaly detector. This paper studies a semi-supervised anomaly detector based on support vector machines, which takes the best of existing supervised and unsupervised support vector machines algorithms. The proposed algorithm allows the maximum proportion of vectors detected as anomalies and the maximum proportion of errors in the supervised data to be controlled, through two hyperparameters defining these proportions. Simulations conducted on various benchmark datasets show the interest of the proposed semi-supervised anomaly detection method.