Autonomous Identity Verification and Behavioral Anomaly Detection for Maritime Surveillance: An Edge-Deployable Pipeline and Simulation Study

The Automatic Identification System (AIS) is unauthenticated, so vessels can broadcast false identities or conceal suspicious behavior. We describe a lightweight, unsupervised pipeline intended to run on a shipboard computer without a shore or cloud link. It aggregates a week of AIS reports into a behavioral profile per vessel, applies a deterministic hydrodynamic rule, scores profiles with an Isolation Forest, and attaches a plain-language triage note to every alert. On a transparent simulator with four injected anomaly types (5,000 vessels, 5 seeds), class-conditional features raised the Isolation Forest's F1 score from 0.49 to 0.71 and its recall on size-identity cloaking from 0.37 to 0.93. Contrary to our initial hypothesis, exponential-moving-average smoothing of speed reduced performance. The model exports to ONNX with identical decisions and scores about 13-22 microseconds per vessel in batch on a two-core CPU. All results are on synthetic data; validation on real AIS data and embedded hardware is future work. This record contains the preprint (not peer reviewed) and the complete code to reproduce every table and figure.