China Coastal Bulk (Coal) Freight Index Forecasting Based on an Integrated Model Combining ARMA, GM and BP Model Optimized by GA
Round 1
Reviewer 1 Report
Overall the research is a timely and important one to explore. The study has good contribution potential. However, the paper's content should be checked with a professional proofreader. There are a few grammatical and punctuation errors that, if corrected, will enhance the quality of the paper.
Author Response
Point 1: There are a few grammatical and punctuation errors that, if corrected, will enhance the quality of the paper.
Response 1: We have read the entire article carefully and corrected the grammatical and punctuation errors in the paper.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Overall this research article is interesting. Some points need to highlights by authors:
1) Contribution of this research is not clear.
2) Research work flow is complex. Try to make it simple.
3) Comparison analysis need to be performed that shows the comparison of proposed method with existing methods.
Author Response
Point 1: Contribution of this research is not clear.
Response 1: In lines 473-495 of the paper, we have clearly divided the contributions of this paper into four points and elaborate on each of them.
Point 2: Research work flow is complex. Try to make it simple.
Response 2: In this paper, the working steps of the combinatorial model construction are clearly written in lines 276-294. Also, you can see the workflow of this paper according to Figure 1, which is not complicated, and it is a common idea to build a combinatorial model.
Point 3: Comparison analysis need to be performed that shows the comparison of proposed method with existing methods.
Response 3: In lines 440-456, we compare the MAE, RMSE, TIC, and AE values of the ARMA-GM-GABP combined model, ARMA model and GM (1,1) model. The comparative analysis is then performed based on the obtained results.
Author Response File: Author Response.docx
Reviewer 3 Report
The article is at an early stage and needs significant revision and improvement. I will not describe the problems and necessary changes in sections 1-2, since the main problem is in section 3.
In my opinion, the main problem of the article are:
I - not matching the standard requirements for structure of the article,
II - poor organization and presentation of the research results.
I. In fact, the article lacks the main components:
- the reasons motivating the authors to this study, that is, why it is so important to have a forecast of the CBCFI index
- knowledge gap,
- novelty in comparison with similar studies,
- possible directions of practical application of the results of the study.
All sections of the article require significant changes, perhaps except for the Methods section.
II. Main questions for the Results section.
Lines 278-282. The authors do not indicate the exact amount of data in the training and test data sets. If looking at Figure 2, the number of points is approximately 30, which is clearly not enough to train the neural network. For any kind of neural networks, the rule is that the more data, the better.
Line 364. "The number of nodes in the input layer is 2 ...."
It must be clearly shown which variables were used in the input layer. Figure 1 shows that the variables were the CBCFI index forecast data obtained by the ARMA and GM methods.
The classical task of forecasting, which is solved using neural networks of this type, is the problem of nonlinear regression.
In fact, the neural network is trying to build a function y = f (x1, x2 ... xn), where x1, x2 ... xn are input variables, y is the output value.
In this case, the authors supply the same variable (CBCFI index) as the input of the network, which they receive at the output.
I don't understand how this network works, please explain with references to scientific sources. Perhaps I missed something, since the science of neural networks is very dynamic.
Lines 415 - ... To assess the quality of the forecast, the authors use 4 formulas. At their core, these formulas are different options for calculating the error, that is, the difference between the forecast and the actual value of the index. That is, there is no logical sense in such calculations and their comparison.
As a rule, the assessment of the quality of forecasting by the network is carried out by calculating and comparing the "error" - "correlation" pair.
And the last. The title of the article stated by the authors is "China Coastal Bulk (Coal) Freight Index Forecasting...", but there is no forecasting!!! It seems to me that it would be logical to end the article using the proposed method to prediction CBCFI index on new input data that the network has not yet "seen".
Author Response
Point I-1: The reasons motivating the authors to this study, that is, why it is so important to have a forecast of the CBCFI index.
Response I-1: We have now elaborated on the motivation for this study in more detail, which you can see on lines 45-50.
Point I-2: Knowledge gap.
Response I-2: We present the current status for CBCFI prediction in lines 45-50, and analyze the problems of existing research in detail in lines 64-77. Based on these unresolved problems, we present the research content and research objectives of this paper in lines 78-83.
Point I-3: Novelty in comparison with similar studies.
Response I-3: We evaluate the previous literature and the similar studies’ limitations in lines 64-77. Then, the first novelty of this paper is mentioned in lines 78-79: using the combined model to predict CBCFI, which can get better predict value. The second novelty of this paper is mentioned in lines 88-90: using genetic algorithm to optimize BP neural network, which can avoid the defects of BP model and improve the combined model's prediction accuracy.
Point I-4: Possible directions of practical application of the results of the study.
Response I-4: We have proposed the practical application direction in lines 492-496, 507-510.
Point II-1: Lines 278-282. The authors do not indicate the exact amount of data in the training and test data sets. If looking at Figure 2, the number of points is approximately 30, which is clearly not enough to train the neural network. For any kind of neural networks, the rule is that the more data, the better.
Response II-1: We mention the data information in lines 295-300. The amount of data in the training data set is 2038, which is sufficient for the training of the neural network model.
Point II-2: "The number of nodes in the input layer is 2 ...." It must be clearly shown which variables were used in the input layer. Figure 1 shows that the variables were the CBCFI index forecast data obtained by the ARMA and GM methods.
Response II-2: We present the input data and output data of the GA-BP neural network in detail when describing the constructing steps of the combinatorial model in section 2.4. You can find it in lines 280-285.
Point II-3: The classical task of forecasting, which is solved using neural networks of this type, is the problem of nonlinear regression.
In fact, the neural network is trying to build a function y = f (x1, x2 ... xn), where x1, x2 ... xn are input variables, y is the output value.
In this case, the authors supply the same variable (CBCFI index) as the input of the network, which they receive at the output. I don't understand how this network works, please explain with references to scientific sources. Perhaps I missed something, since the science of neural networks is very dynamic.
Response II-3: The principles of combinatorial model construction are described in detail in lines 257-272. In the formula , where (i=1, 2, …, n) represents the prediction results of -th prediction methods, and is a nonlinear function that performs a nonlinear combination of the prediction results obtained by different prediction methods. Since a single hidden layer BP network can arbitrarily approximate a continuous nonlinear function, this paper attempts to use BP neural networks to model the nonlinear combinatorial prediction function , so as to achieve the purpose of nonlinear combinatorial modeling and prediction using ARMA model and GM (1, 1) model. Thus, the input data of the GA-BP model are the predicted values of the ARMA model and the GM (1, 1) model, and the output data are the predicted values of the combined model generated by the GA-BP neural network after the nonlinear combination of these two predicted values.
Point II-4: ... To assess the quality of the forecast, the authors use 4 formulas. At their core, these formulas are different options for calculating the error, that is, the difference between the forecast and the actual value of the index. That is, there is no logical sense in such calculations and their comparison. As a rule, the assessment of the quality of forecasting by the network is carried out by calculating and comparing the "error" - "correlation" pair.
Response II-4: We have added an explanation of the role of the four evaluation indicators in lines 429-437 and explained why they were chosen.
Point II-5: The title of the article stated by the authors is "China Coastal Bulk (Coal) Freight Index Forecasting...", but there is no forecasting!!! It seems to me that it would be logical to end the article using the proposed method to prediction CBCFI index on new input data that the network has not yet "seen".
Response II-5: In Section 3.5, we forecast the November 2019 CBCFI data using the ARMA model, the GM model and the combined ARMA-GM-GABP forecasting model, and compare the forecast values with the true values. We present the forecasting results of the three models through Figures 4, 5, and 6. It can be seen that our proposed forecasting model is able to predict the CBCFI values more accurately, and for the volatile part of the CBCFI, it can also accurately fit the trend.
Author Response File: 
Author Response.pdf
Reviewer 4 Report
I have the following recommendations regarding the paper with a title China Coastal Bulk (Coal) Freight Index Forecasting Based on an Integrated Model Combining ARMA, GM and BP Model Optimized by GA.
1. Paper does not evaluate the previouse literature properly with limitations of the previouse works and research gap.
2. Authors should mention the novelty motivation
3. The contribution points and problem statment should be clearly described in the paper.
4. Regarding the research design. Some flow charts or diagrams need be inserted in the paper.
5. Results should be improved with proper description, discussion and comparison with other related works.
6. Authors should mention some future research directions
7. A thorough proof read is required for checking the english, typos, grammatical mistakes and sentences structure.
8. Avoid repetetive words and sentences.
9. Improve the figures
Author Response
Point 1: Paper does not evaluate the previous literature properly with limitations of the previous works and research gap.
Response 1: After analyzing the previous literature, we have now re-written the evaluation of the previous literature and the work limitations and research gaps in lines 64-77.
Point 2: Authors should mention the novelty motivation.
Response 2: We have elaborated on the motivation for this study in more detail, which you can see it in lines 45-50.
Point 3: The contribution points and problem statement should be clearly described in the paper.
Response 3: We have explicitly described the contribution points on lines 473-495 and mentioned the problem statement on lines 78-83.
Point 4: Regarding the research design. Some flow charts or diagrams need be inserted in the paper.
Response 4: We clearly depict the research design flow of the ARMA-GM-GABP combined model in Figure 1.
Point 5: Results should be improved with proper description, discussion and comparison with other related works.
Response 5: In lines 439-455, we describe and compare the MAE, RMSE, TIC, and AE values of the combined model, ARMA model and GM (1,1) model. The comparative analysis and discussion are then performed based on the obtained results.
Point 6: Authors should mention some future research directions.
Response 6: We have proposed the future research directions in lines 503-509.
Point 7: A thorough proof read is required for checking the English, typos, grammatical mistakes and sentence structure.
Response 7: We have carefully checked and proofread the English, typos, grammatical mistakes and sentence structure in the paper.
Point 8: Avoid repetitive words and sentences.
Response 8: We have carried out a detailed inspection of the repeated statements in the article and removed the repeated statements.
Point 9: Improve the figures
Response 9: We have seriously improved the chart.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
I recommend accepting the article in present form.
Reviewer 4 Report
The authors have addressed all the comments.
