ABiLSTM Based Prediction Model for AUV Trajectory

Round 1

Reviewer 1 Report

I recommend for the major revision, here I have mentioned some of my concerns. 

1. The abstract has to improve, and it should be trimmed. The author should explain the key results in the abstract. 

2. To correct grammatical errors, it should be read by someone with good English and grammatical errors should be corrected. 

3. It is necessary to increase and update the number of references by adding some latest references. The authors should cite some of the recent 2022-2023 articles. 

4. Texts present in Fig. 1,2, and 4 are not clear 5. The conclusion can be further improved. 

6. Based on your expertise and knowledge, do you have any suggestions or recommendations for improving the proposed model or for further research in this area? 

7. Why Fig. 10 is shown in black and white 

8. All equations should be cited properly 

9. The quality of Fig.11 has to be improved, improving resolution and size 

10. Is there a proper comparison with other relevant models such as LSTM, BiLSTM, and Attention-LSTM?

 

The English should be polished further.

Author Response

See attachment, please.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article is focused on prediction of trajectory of marine vessels. The authors use LSTM neural network with attention mechanism. In the paper the authors discuss their solution on two examples: AUV (automated underwater vessel) and ships data from AIS dataset. The paper is in general well written, but some issues have to be repaired:

-The authors show errors of their prediction model. Unfortunately, it is not obvious if figures 11 and 12 presents train and test data samples or only one of these sets. Moreover, the authors did not provide the information is comparison of the models (table 2) is done on both sets or only on test set. This distinction is very important to assess the models and select the best one.

-What is the training time difference between the models? Is the increase of accuracy worth the time.

-It sounds like the authors tested model on single event of AUV. In real life case the proof of the model should be done with use of more than one event. Probably the model needs to be trained on each event individually.

-In the case of recovery of lost AUV, the most important information is the final position of the vessel, probably on ground level, did authors tried to assess the inacurrance of this position. As is written the altitude is the worse predicted parameter of all. If the final position cannot be predicted with acceptable error, how the proposed method can be used in real-life scenario of lost vessel recovery? How the recovery scenario with use of the model look like?

-On what base the parameters of the network were chosen (tables 1 and 3). Did authors tried different settings? What were the results of such changes?

-When compare two independent prediction system the total loss measures like RMSE/MSE/MAE are not enough. The authors should also give information on the statistics of errors. Probably the rank test (like Wilcoxon rank test) would prove that the predictions of ABiLSTM predictions are statistically better than the others.

-The images 11, and 12 are too small and are not readable at all.

-Where on images 11, 12 we have training and test samples? Can they be visually distinguished?  

 

Author Response

See attachment, please.

Author Response File: Author Response.pdf

Reviewer 3 Report

The topic & way the authors presented the work is really good, but here i've few suggestions to the authors to improve the quality of the manuscript.

1. The introduction part should contain related work，and it needs to reduce the information unrelated to this work.

2. In section 2.4, how many and which kinds of attention mechanism is used should be clarified. you may explore various attention mechanism & cite a reference in section 2.4:

3. The words in results are too small & invisible.

4. the authors need to provide a comprehensive explanation of how each model was implemented, the paper does not provide a clear description of the dataset used for training and testing, which is crucial information for the readers to evaluate the validity of the results. 

5. provide a compartive table in the end of the manuscript before conclusion to depict the focus and meaning of this study.

 English language fine, Kindly check for spell mistakes & grammatical mistakes.

Author Response

See attachment, please.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The author made sufficient changes and the article can be accepted in the current form. 

Reviewer 3 Report

Thanks to authors for addressing all issues raised by my end. Slight grammatical & Spell check is required.

Editing is required