Anomaly detection of trajectories with kernel density estimation by conformal prediction

Abstract

This paper describes conformal prediction techniques for detecting anomalous trajectories in the maritime domain. The data used in experiments were obtained from Automatic Identification System (AIS) broadcasts-a system for tracking vessel locations. A dimensionality reduction package is used and a kernel density estimation function as a nonconformity measure has been applied to detect anomalies. We propose average p-value as an efficiency criteria for conformal anomaly detection. A comparison with a k-nearest neighbours non-conformity measure is presented and the results are discussed.