Fitting Empirical Distributions for Vessels Behavioural Analysis and Maritime Anomaly Detection

Naval behaviour characterization and maritime anomaly detection represents two active area of research. Although multiple probabilistic approaches have been implemented for explaining different parametric maritime phenomena, most of them are focused only on the analysis of primary kinematic data, such as position, speed, course of vessels, or lateral displacement relative to an obstacle. This article presents an analysis on multiple complex maritime activities (e.g., fishing, drifting, hidden activities) that were derived by processing AIS messages collected in the Black Sea region. Quantitative histograms were made for each type of activity while probability and cumulative distribution functions were plotted using kernel density estimation applications. After that, multiple evaluation metrics analysed the possibilities of representing these distributions in the form of known parametric distributions (e.g., Gaussian, Gamma, Rayleigh, Log-normal). Finally, there were extracted the parameters that best fitted these distributions. Also, a universal parametric distribution was searched to best fit all observed maritime activities.