A Cloud-based Approach for Ship Stay Behavior Classification using Massive Trajectory Data

With the widespread application of AIS (Automatic Ship Identification System), ship trajectory data is being collected and becoming increasingly available. Consequently, a lot of ship trajectory data applications have become feasible that mine the value from the data. In this paper, based on massive ship trajectory data, we aim to classify two kinds of ship stay behavior for recognizing different areas in the port, namely berth and anchorage. The traditional trajectory data classification model mainly distinguishes the moving and staying state of moving objects, but there is little research on the classification of different kinds of stay behavior, especially for ship stay behavior classification. In this work, we propose an extraction algorithm based on the cloud storage and distributed computing frameworks to extract classification features by analyzing the behavioral characteristics of ships at berths and anchors. Second, with the consideration of the low precision, drift and sparsity characteristics of ship trajectory data, we design a series of experiments based on ten-fold cross-validation method for evaluating five classical classification models, such as XGBoost, Random Forest and so on. Third, experimental verifications of various classification models are conducted based on a real ship trajectory dataset, and the effectiveness of different models for recognizing ship stay area are compared.