Distance metric learning for conditional anomaly detection

Anomaly detection methods can be very useful in iden-tifying unusual or interesting patterns in data. A re-cently proposed conditional anomaly detection frame-work extends anomaly detection to the problem of iden-tifying anomalous patterns on a subset of attributes in the data. The anomaly always depends (is conditioned) on the value of remaining attributes. The work pre-sented in this paper focuses on instance–based meth-ods for detecting conditional anomalies. The methods depend heavily on the distance metric that lets us iden-tify examples in the dataset that are most critical for de-tecting the anomaly. To optimize the performance of the anomaly detection methods we explore and study metric learning methods. We evaluate the quality of our methods on the Pneumonia PORT dataset by de-tecting unusual admission decisions for patients with the community–acquired pneumonia. The results of our metric learning methods show an improved detection performance over standard distance metrics, which is very promising for building automated anomaly detec-tion systems for variety of intelligent monitoring appli-cations