Integrated Detection Of Ship Abnormal Behavior Based On Isolation F Orest Algorithm

This paper aims to assist the maritime department to monitor waters and identify more meaningful outliers.In order to assist the managers to identify the anomaly behavior of ships,which improve the efficiency of regulation.To detect the abnormal trajectory and the point of moving ship,offering suggestions of transportation planning and management.We preprocess the AIS data,simplify the trajectory of moving ships and use DBSCAN algorithm to segment and cluster the motion trajectory,artificial select the exception classes of the results of DBSCAN algorithm and determine the anomaly trajectory which far away from the center of clustering.Based on the raw data provided by Automatic Identification System in the normal class,six groups of data that can represent the motion characteristics of ships were figured out.The iForest algorithm was introduced into the field of ship abnormal behavior detection.The single abnormal scores were calculated by using the.iForest algorithm model and the integrated scores were calculated by Integrated Analysis and Entropy Method.The test set’s evaluation indicators were used to set the judgment standard and AIS data was imported to verify the effectiveness of the algorithm.The results show that the detection rate of outliers is 100%.The precision rate and recall rate are greatly influenced by the selection of threshold.Thealgorithm has good performance on ROC curve,the value of AUC is 0.9883.This model can be applied to perform the real time monitoring towards the ship abnormal behaviors,make up for the lack of the single detection results and judge the types of abnormalities so as to highly improve the supervision efficiency of maritime.And the model combine the interaction between ships with the detection of ship abnormal behavior so as to make it more suitable and comprehensive for massive data.The main work of this paper is as follows:(1)Review and classify the research of detection for ship abnormal behavior by the results of their research.We have found problems that the result of detection is single,the interaction of ship is not considered and the high rate of false alarm.(2)A method for the detection of ship abnormal behavior is presented which combine the DBSCAN algorithm with iForest algorithm.The results of ship abnormal behavior are divided into eight categories clearly,such as trajectory,speed,course,rate of turn,relative distance,relative velocity and others.(3)The iForest algorithm is first applied in marine traffic.We discuss the feasibility of the frame for detection which judge the abnormal behavior by the characteristics of data instead of establishing a model of normal behaviors.