Using new attribute construction to incorporate the expertise of human experts into a smuggling vessels classification system

摘要

Modifying the database originally designed for maritime transportation and harbourside sailing safety management in order to train the data classifier to detect smuggling and illegal immigrant vessels may lead to a lack of class-related attributes, and the accuracy rate of classification may be low. However, some valuable artificial classification rules already exist prior to the building of a data classifier, and it is worth attempting to incorporate them into the database and thus build a more accurate data classifier. With the limiting condition that no extra reference point of the terrain can be added into the system, this research selects the artificial smuggling detection rule “vessels sail fast towards shore” deduced by Coast Guard’s officers, sets the radar station as the reference point of the shore coordinates, and defines a new attribute, the absolute difference between the anti-course and target azimuth, to demonstrate the “vessels sail approximately towards radar station or sea” concept. Matching the original “speed” attribute, the newly-modified database includes the intrinsic meaning of “vessels sail fast towards radar station or sea”. The results show that the accuracy rate of the two data classifiers, the back-propagation neural network (BPN) and support vector machine (SVM), improved by 22% and 14%, respectively. It is significant that incorporation of the expertise of human experts by new attribute construction contributes to the effective learning of the data classifier.