Transferring the Contamination Factor between Anomaly Detection Domains by Shape Similarity

Anomaly detection attempts to find examples in a dataset that do not conform the expected behavior. Algorithms for this task assign an anomaly score each example representing its degree of anomalousness. Setting threshold on scores enables converting these into discrete prediction example. appropriate is challenging practice since often treated as unsupervised problem. A common approach set based dataset's contamination factor, i.e., proportion anomalous data. While factor may be known domain knowledge, it necessary estimate by labeling However, many problems involve monitoring multiple related, yet slightly different entities (e.g., fleet machines). Then, estimating separately data would extremely time-consuming. Therefore, paper introduces method transferring from one (the source domain) related where unknown target domain). Our does require labeled and modeling shape distribution both domains. We theoretically analyze how our behaves when (biased) converges true one. Empirically, outperforms several baselines real-world datasets.