Identification of Spoofing Ships from Automatic Identification System Data via Trajectory Segmentation and Isolation Forest

Round 1

Reviewer 1 Report

Summary/Contribution: The work proposes pre-processing Automatic Identification System (AIS) data to accurately identify and remove ship trajectory outliers, particularly spoofing ships with unauthorized maritime mobile service identification code (MMSI) owned by normal ships. Outliers in ship trajectory from AIS can influence marine situation awareness, and earlier approaches of deleting trajectory points of spoofing ships lose crucial information. Trajectory characteristics mining reveals changes in speed and distance between consecutive trajectory points between normal ship trajectory and normal ship trajectory mixed with spoofing ship. The authors suggest trajectory segmentation by update time interval threshold and isolated forest to train labeled trajectory points of regular ships mixed with spoofing ships to identify ships with lengthy update time intervals. The experimental findings show that the proposed method can practically eliminate trajectory outliers and identify spoofing ships with an average accuracy of 88.4% to 93.3%, depending on trajectory segmentation by update time interval.
Comments/Suggestions:

1. The research challenge and its impetus could be stated more clearly and concisely in the introduction. Why is there a need for a new approach to identifying spoofing ships if the existing literature on AIS data quality already has solutions to the problems it identifies?

2. More specifics on the approach for mining of trajectory characteristics from AIS data, such as the attributes used to differentiate between regular and spoofing ships, would improve this section.

3. More detail on the data source used in the study, such as its extent in terms of geography and time, as well as its collection and curation methods, would be beneficial.

4. Additional details on the trajectory segmentation process utilized in the study, such as the criteria used to calculate the time interval threshold between trajectory points, would be beneficial.

5. Section 2 is too short and needs to be extended.

6. Section 3.2. could benefit from more detailed information on the trajectory segmentation process, including the specific criteria used to determine the time interval threshold between trajectory points, and how this threshold is used to segment the trajectories into smaller segments for analysis.

7. Additionally, it would be helpful to provide more information on how missing or jumping trajectory points are handled during the segmentation process, and how this affects the accuracy of the results.   8. The schematic overview provided in Figure 2 is helpful, but it could benefit from more detailed annotations and explanations of the various components of the framework, including how they are related to each other and how they contribute to the overall accuracy of the proposed approach.
9. Additionally, it would be useful to provide more information on how the various parameters and thresholds used in the approach are determined, and how they can be tuned to optimize performance for different settings and scenarios.

10. The authors are invited to add a new paragraph about the use of runtime testing for validating the proposed approach. For this purpose, the authors may include the following interesting references (and others):

a. https://www.inderscienceonline.com/doi/abs/10.1504/IJCAT.2015.070489

b. https://ieeexplore.ieee.org/abstract/document/6595504

can be improved

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear authors,

The content and subject matter are interesting, but I don't know if it's good enough in terms of academic contribution.

There are rather few related studies, and it is difficult to say that it is a new study other than data.

Q1. It's hard to know what data Figure 1 represents.

Q2. There is no reason why Isolation Forest was chosen among the many unsupervised learning methods for anomaly detection. Is it related to that data?

Q3. The trajectory characteristics of the normal ship and the spoof ship described in the paper cannot be known by looking at Figure 3.

Q4. The description of the correlation between the continuity of the Class 4 spoof ship trajectory and the low accuracy of the model needs to be supplemented.

Also, there are typos in some parts, such as equations, so please fix them.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors considered my comments and suggestions.

 

Can be improved

 

Reviewer 2 Report

The reviewer has confirmed all Comments have been made to the paper presented.

The reviewer suggests publishing without any changes.