Michael Davis (CERN)
Often we have a dataset where most of the data is produced by a known mechanism (or several mechanisms) which we understand, but some data is produced by a different process. In such cases we can consider the known processes as background to detect the signal. However, what about the case where we do not have knowledge about the underlying processes, but want to detect which part of our dataset is unusual or anomalous? One method that can be used in this case is density-based outlier detection, where each data point is considered in relation to its local neighbourhood.