Prediction of Construction and Production Safety Accidents in China Based on Time Series Analysis Combination Model
Round 1
Reviewer 1 Report
The paper deals with prediction of safety accidents and deaths in construction industry in China. The authors consider various models to choose the ones that show better results. The overall impression is positive. As a suggestion I recommend to focus more on data preprocessing techniques and their impact on the quality of prediction. In particular, how the facts mentioned in lines 125-137 can be used in predicting future global trends.
Author Response
Dear reviewer,
We really appreciate you for your carefulness and conscientiousness. Your comments are really valuable and helpful for revising and improving our paper. We have also asked a professor who is a native English speaker to correct the details and made meticulous modifications to this manuscript based on your proposal. According to your suggestions, point-by-point responses are listed as follows.
Point 1: The paper deals with prediction of safety accidents and deaths in construction industry in China. The authors consider various models to choose the ones that show better results. The overall impression is positive.
Response 1: First, thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits. We've also made grammar and vocabulary fixes which were all highlighted by using yellow colored text.
Point 2: As a suggestion I recommend to focus more on data preprocessing techniques and their impacts on the quality of prediction.
Response 2: Thank you very much for your constructive comments. We think it is important to highlight your description of data preprocessing techniques and their impacts on the quality of predictions. This article preprocesses the data by using the trend decomposition method. But as you can see, the rationality of this method is not clearly stated, so we have added the literature support of this preprocessing technique in the introduction and the positive role played in our study. (For details, you can see page 2, line 55-67; page 3, line106-112)
Point 3: In particular, how the facts mentioned in lines 125-137 can be used in predicting future global trends.
Response 3: Thank you very much for your valuable question. Regarding whether the analysis you mentioned about the development trend of construction accidents in China mentioned by line125-137 can be used to predict future global trends, we think your proposal is very valuable. Because through our long-term research on the macro prediction of accidents, we have found that the construction industry or other industries in each country has its own unique emergencies, and we should consider those special circumstances when we study. This means that the addition of seemingly unrelated indicators or factors may have an impact on accident prediction. This is one of the reasons why the nature of accidents in my paper is considered when predicting the number of deaths, quantifying the impact of larger and above accidents has proved to be necessary. We've added the sections you've marked.(For details, you can see page 4, line 142-149)
Once again, thank you very much for your comments and suggestions. And we hope that the corrections will meet with approval.
Reviewer 2 Report
-
Author Response
Dear reviewer,
We really appreciate you for your carefulness and conscientiousness. Your comments are really valuable and helpful for revising and improving our paper. According to your comments, point-by-point responses are listed as follows.
Point 1: The research design is not applicable.
Response 1: First, thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits. I'm so sorry about your discontent with the design of my research. We think there must be something to improve in our article. After you did not give a specific opinion, we still made the following changes.
1.About data preprocessing techniques and their impacts on the quality of prediction
We think it is important to highlight your description of data preprocessing techniques and their impacts on the quality of predictions. This article preprocesses the data by using the trend decomposition method. But as you can see, the rationality of this method is not clearly stated, so we have added the literature support of this preprocessing technique in the introduction and the positive role played in our study. (For details, you can see page2 line 57-69; page 3, line108-117)
2.How the facts mentioned about figure1 can be used in predicting future global trends
Through our long-term research on the macro prediction of accidents, we have found that the construction industry or other industries in each country has its own unique emergencies, and we should consider those special circumstances when we study. This means that the addition of seemingly unrelated indicators or factors may have an impact on accident prediction. This is one of the reasons why the nature of accidents in my paper is considered when predicting the number of deaths, quantifying the impact of larger and above accidents has proved to be necessary. We've added the sections you've marked.(For details, you can see page 4, line 147-154)
3.Lacks more emphasis on the direct application of the results
We have made a small amount of clarification in the “conclusion” section about why the results and true values are quite different and not very informative due to the impact of the epidemic that began in 2020. But because the core of our paper is time series analysis and processing and the construction of a combined prediction model, the prediction results have been well demonstrated before the epidemic. However, due to our insufficient expression, it may lead to the reader's confusion about the results. Based on your suggestions, we have added additional content to emphasize the feasibility and reliability of our model and the prospects for future development.(For details, you can see page 17, line556-568, line 573-579)
4.The appropriate size, scale for the implementation has not been discussed clearly
First, the data comes from China's housing municipal engineering safety production accidents released by the Ministry of Housing and Urban-Rural Development in China of which the number of monthly accidents and the number of deaths from 2009 to 2019 are recorded in detail. So we use these data as the original data for the time series, and the reason why we do not consider the data before and after the interval is based on the development of China's construction industry and the impact of the COVID-19 epidemic, respectively.(For details, you can see page page 5, line 180-185)
Secondly, we applied it to the construction field after the model was built, but the model itself can basically be applied to various countries and fields around the world. Because we preprocess and forecast the data as a time series, there is no special limit to the size of the data itself. And the data processed in the article is limited because only accurate data for these 11 years can be obtained at this stage. Therefore, in future research, we can continue to use the results of this research to predict the accident situation in different industries or further refine it to optimize the model.(For details, you can see page2, line 57-69; page 17, line 556-568)
Finally, we want to add more indicator factors in the future, including all levels of society. We believe that those factors that are strongly related to accidents must be the trend that affects the development of accidents, but those seemingly unrelated factors often affect the volatility of the trend. So we think it's necessary to integrate all kinds of indicators into a semi-open system in order to constantly supplement the indicators related to forecasting, and we are currently working on this, which will be reflected in our future work.(For details, you can see page 4, line 147-154; page 17, line 573-579)
We have added many explanations of parameters, values and objects to facilitate the reader's better understanding. We've also made grammar and vocabulary fixes which were all highlighted by using yellow colored text. Once again, thank you very much for your comments and suggestions. We hope that the corrections will meet with approval.
Author Response File: 
Author Response.pdf
Reviewer 3 Report
This paper focuses on modelling and prediction in construction and production safety accidents in China. The article lacks more emphasis on the direct application of the results, which would have made the presentation of the scientific approaches used in addressing the stated objective more attractive. The results can be helpful in the practice of prevention and reduction of construction and production safety accidents.
No misconduct was found within the criteria observed. I have no comments on the paper.
Author Response
Dear reviewer,
We really appreciate you for your carefulness and conscientiousness. Your comments are really valuable and helpful for revising and improving our paper. We have also asked a professor who is a native English speaker to correct the details and made meticulous modifications to this manuscript based on your proposal. According to your suggestions, point-by-point responses are listed as follows.
Point 1: This paper focuses on modelling and prediction in construction and production safety accidents in China. The results can be helpful in the practice of prevention and reduction of construction and production safety accidents. No misconduct was found within the criteria observed. I have no comments on the paper.
Response 1: First, thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits.
Point 2: The article lacks more emphasis on the direct application of the results, which would have made the presentation of the scientific approaches used in addressing the stated objective more attractive.
Response 2: We have made a small amount of clarification in the “conclusion” section about why the results and true values are quite different and not very informative due to the impact of the epidemic that began in 2020. But because the core of our paper is time series analysis and processing and the construction of a combined prediction model, the prediction results have been well demonstrated before the epidemic. However, due to our insufficient expression, it may lead to the reader's confusion about the results. Based on your suggestions, we have added additional content to emphasize the feasibility and reliability of our model and the prospects for future development.(For details, you can see page 17, line556-568, line 573-579)
Once again, thank you very much for your comments and suggestions. And we hope that the corrections will meet with approval.
Author Response File: Author Response.pdf
Reviewer 4 Report
It's an interesting research on the Prediction of construction and production safety accidents.
As the authors mentioned, the purpose of this study is to use trend decomposition method to reduce the fluctuation of non-stationary time series, and use the combination of Autoregressive Integrated Moving Average model and Grey model with the frational order accumulation to accurately predict the construction accidents.
However,.the appropriate size, scale for the implementation has not been discussed clearly.
.
The discussion and conclusion can be improved.
Author Response
Dear viewer,
Response to Reviewer 4’s Comments
We really appreciate you for your carefulness and conscientiousness. Your comments are really valuable and helpful for revising and improving our paper. We have also asked a professor who is a native English speaker to correct the details and made meticulous modifications to this manuscript based on your proposal. According to your suggestions, point-by-point responses are listed as follows.
Point 1: It's an interesting research on the Prediction of construction and production safety accidents. As the authors mentioned, the purpose of this study is to use trend decomposition method to reduce the fluctuation of non-stationary time series, and use the combination of Autoregressive Integrated Moving Average model and Grey model with the fractional order accumulation to accurately predict the construction accidents.
Response 1: First, thank you very much for your time involved in reviewing the manuscript and your very encouraging comments on the merits.
Point 2: However, the appropriate size, scale for the implementation has not been discussed clearly.
Response 2: I'm sorry, but I'm not sure if we can understand that we didn't explain clearly in the article why we used the data from 2009 to 2019, the applicable boundaries or extents of the model, or the size requirements for the data. Based on your suggestions, we have added additional content and our manuscript were all highlighted by using yellow colored text.
First, the data comes from China's housing municipal engineering safety production accidents released by the Ministry of Housing and Urban-Rural Development in China of which the number of monthly accidents and the number of deaths from 2009 to 2019 are recorded in detail. So we use these data as the original data for the time series, and the reason why we do not consider the data before and after the interval is based on the development of China's construction industry and the impact of the COVID-19 epidemic, respectively.(For details, you can see page 5, line 180-185)
Secondly, we applied it to the construction field after the model was built, but the model itself can basically be applied to various countries and fields around the world. Because we preprocess and forecast the data as a time series, there is no special limit to the size of the data itself. And the data processed in the article is limited because only accurate data for these 11 years can be obtained at this stage. Therefore, in future research, we can continue to use the results of this research to predict the accident situation in different industries or further refine it to optimize the model.(For details, you can see page 2, line 57-69; page 17, line 556-568)
Finally, we want to add more indicator factors in the future, including all levels of society. We believe that those factors that are strongly related to accidents must be the trend that affects the development of accidents, but those seemingly unrelated factors often affect the volatility of the trend. So we think it's necessary to integrate all kinds of indicators into a semi-open system in order to constantly supplement the indicators related to forecasting, and we are currently working on this, which will be reflected in our future work.(For details, you can see page 4, line 147-154; page 17, line 573-579)
Once again, thank you very much for your comments and suggestions. And we hope that the corrections will meet with approval.
Author Response File: Author Response.pdf
Reviewer 5 Report
The authors say in line 14 that: "use the combination of Autoregressive Integrated Moving Average model and Gray model with the fractional order accumulation to accurately predict the construction accidents." If accidents could be predicted, would that mean they could be avoided?
Numerous phrases or paragraphs must be reformulated, and/or given additional explanations on the coefficients or terms used, for example: lines 50-52, 101, 161, 189-190, 205-208, 228-234, 279 all the paragraph 2.2.
The models used must be explained in more detail: ARIMA and Gray, what is the difference between the two. What does "the rolling forecast method" mean?
Why is the research stopped at the level of 2019 and not updated until now?
On lines 136, 136 the authors say: "and the soaring housing prices in some parts of the country", it is not clear what the connection would be with the proposed topic.
The terms "Trend, Seasonal and Residual" used in figure 2 must be detailed, and the corresponding graphs explained. Why does the Seasonal graph have an identical annual fluctuation? What do the values on the abscissa mean and what units do they have? Horizontally, the breakdown of accidents and deaths was expressed annually and not monthly.
The authors say on line 168 "Among the three data sets" what are the three data sets or where are they found in the article?
The authors say on line 189 "By observing the Autocorrelation Function (ACF) and partial Autocorrelation Function (PACF) plot" where are these graphs? Why choose the ARIMA model? (line 191)
The notations used in all formulas 1-6 must be explained.
Figure 3 must be explained. What is the connection with the researched subject?
How were the values from table 2 and those from figure 4 obtained. Figure 4 is a table.
The terms: "dynamic prediction" and "static prediction" (line 279) must be explained. How were these terms determined at line 284 "dynamic prediction is 0.36" and at line 285 "static predicted....of 0.20"?
Table 1 on lines 312-329 is entered incorrectly.
Author Response
Dear reviewer,
We really appreciate you for your carefulness and conscientiousness. Your comments are really valuable and helpful for revising and improving our paper. We have also asked a professor who is a native English speaker to correct the details and made meticulous modifications to this manuscript based on your proposal. According to your suggestions, point-by-point responses are listed as follows.
Point 1: The authors say in line 14 that: "use the combination of Autoregressive Integrated Moving Average model and Gray model with the fractional order accumulation to accurately predict the construction accidents." If accidents could be predicted, would that mean they could be avoided?
Response 1:Yes,it is. Accident can be avoided because it has certain characteristics and laws, as long as these characteristics and patterns are mastered, and can be reasonably applied, effective measures can be taken to control them in advance, and the occurrence of accidents and the losses caused by them can be prevented and reduced. It is by analyzing these characteristics and patterns that predictions support future safety efforts.
Point 2: Numerous phrases or paragraphs must be reformulated, and/or given additional explanations on the coefficients or terms used, for example: lines 50-52, 101, 161, 189-190, 205-208, 228-234, 279 all the paragraph 2.2.
Response 2: Thank you for your careful reading and patient explanation, we have made relevant changes and additions. (For details, you can see page2-7 , line highlighted in yellow)
Point 3: The models used must be explained in more detail: ARIMA and Gray, what is the difference between the two. What does "the rolling forecast method" mean?
Response 3: â‘ Difference: ARIMA: treats the data series formed by the predicted object over time as a random sequence, differentiates its sequence to make it stable, and then uses a certain mathematical model to approximate the sequence. Once identified, this model can predict future values based on past and present values of the time series. The advantage is that the model only needs endogenous variables and does not need to rely on other exogenous variables. (Rely only on the data itself, unlike regression that requires other variables). The disadvantage is that the time series data is required to be stable or stable by differentiation. Essentially, only linear relationships can be captured, not nonlinear relationships.
Grey: Among many gray theory algorithms, they are often used to make predictions with small samples and less informational data. A method of accumulating (or other processing to generate) the original data to obtain an approximate exponential law and then modeling. The advantage is that it can process less eigenvalue data and do not need a large enough sample space of the data to solve the problems of less historical data, low sequence integrity and reliability, and can generate irregular raw data to obtain a strong regular generation sequence. The disadvantage is that it is only suitable for short- and medium-term forecasts, and only for forecasts that approximate exponential growth.
The reason why the two methods are considered to be effective is based on the fact that there may be insufficient sample data, instability, etc. in the prediction, and the data used in this paper is the case. Combining the advantages of the two models can solve this problem well.
â‘¡Rolling forecasting means that the model is reestimated in each iteration and produces a prediction result. After that, a new observation is added at the end of the series, and the process continues. When there is no more data to add, the process stops. In this paper, the main performance is that for the forecast of annual data, due to the small sample size and large fluctuations of the data, the results are accidental. Therefore, consider using quarterly data with a relatively large sample size, adopt a rolling forecast starting from 2009 according to the forecast every 2 years, every 3 years, and once every 4 years, and then repeat the operation from 2010, and so on, use the results updated after each step to re-forecast, sum the predicted quarterly data, and generate a layered annual forecast result. (For details, you can see page3 , line 108-117 )
Point 4: Why is the research stopped at the level of 2019 and not updated until now?
Response 4:I'm sorry we didn't discussed clearly in the text. The model we built is universal for time series forecasting, but due to the impact of major emergencies such as the COVID-19 epidemic, there is no way to strongly verify the forecast results, but our accuracy of the forecast results for 2019 before the epidemic has been verified. We have added more explanations.(For details, you can see page 5, line 180-185; page 18 conclusion ).
Point 5: On lines 136, 136 the authors say: "and the soaring housing prices in some parts of the country", it is not clear what the connection would be with the proposed topic.
Response 5:Through our long-term research on the macro prediction of accidents, we have found that the construction industry or other industries in each country has its own unique emergencies, and we should consider those special circumstances when we study. This means that the addition of seemingly unrelated indicators or factors may have an impact on accident prediction. This is one of the reasons why the nature of accidents in my paper is considered when predicting the number of deaths, quantifying the impact of larger and above accidents has proved to be necessary. We've added the sections you've marked.(For details, you can see page 4, line 147-154 )
Point 6: The terms "Trend, Seasonal and Residual" used in figure 2 must be detailed, and the corresponding graphs explained. Why does the Seasonal graph have an identical annual fluctuation? What do the values on the abscissa mean and what units do they have? Horizontally, the breakdown of accidents and deaths was expressed annually and not monthly.
Response 6:We have added explanations. The Seasonal graph maintains have an identical annual fluctuation with the occurrence of accidents in each month of the year in China, which is in line with the facts. The units on the abscissa are months, from January 2009 to December 2019, and I'm sorry not to list them all due to image size limitations.(For details, you can see page4 , line 164-175 )
Point 7: The authors say on line 168 "Among the three data sets" what are the three data sets or where are they found in the article?
Response 7:I'm sorry I didn't illustrate clearly. These three sets of data sets are mentioned at the beginning of this paragraph about "summing the values of the long-term trend items obtained by decomposing the quarterly and monthly data trends for accidents and fatalities, and regenerating the annual data" as two sets of data, and the other is the real annual data. Relevant changes and additions have been made in the article.(For details, you can see page 5, line 190-200).
Point 8: The authors say on line 189 "By observing the Autocorrelation Function (ACF) and partial Autocorrelation Function (PACF) plot" where are these graphs? Why choose the ARIMA model? (line 191)
Response 8:I'm so sorry that for reasons of space, we have not included images of the original data in the text, but in response to your comments, we have added explanations of roles of these two images to facilitate the reader's better understanding. And we give this image and related analysis in the subsequent specific application of the model.(For details, you can see page6-7 , line 233-238; page3 , line 108-114 )
Point 9:The notations used in all formulas 1-6 must be explained.Figure 3 must be explained. What is the connection with the researched subject?How were the values from table 2 and those from figure 4 obtained. Figure 4 is a table.
Response 9:We have added more explanations.(For details, you can see page 6-8 , line233-291 ). And We used the Eviews software to perform time series tests. The Figure 4 shows the ACF and PACF plots in the left half and their corresponding test values in the right half.(For details, you can see page 8-9 , line 299-304).
Point 10:The terms: "dynamic prediction" and "static prediction" (line 279) must be explained. How were these terms determined at line 284 "dynamic prediction is 0.36" and at line 285 "static predicted....of 0.20"?
Response 10: We have added explanations. (For details, you can see page 9-10, line 322-332 )
Point 11:Table 1 on lines 312-329 is entered incorrectly.
Response 11: I'm sorry, I've made a correction. (For details, you can see page 10 , line 359 )
Once again, thank you very much for your comments and suggestions. And we hope that the corrections will meet with approval.
Author Response File: Author Response.pdf
Round 2
Reviewer 5 Report
Most of the phrases or paragraphs mentioned (point 2) were not reformulated or given additional explanations.
The authors should include in the article part of the explanations given in the first review (eg points 3, 4, 5).
The authors do not refer in the article to the measures that must be taken or that have been taken to reduce accidents, according to the mathematical study described. The authors should validate the research done with practical evidence to confirm the reduction in the number of accidents. In the introduction, the authors suggest that the mathematical formulas used in the article were also used by other researchers in various fields of activity (plant accidents, traffic accidents, time change of the accident itself, road traffic accidents, production accidents, mine safety production conditions etc.), so the mathematical theoretical part should have already been verified practically. The authors should comment on the practical results and the effectiveness of the method at least through the prism of results from other fields. The conclusions should be completed with those described above.
Author Response
Dear reviewer,
First of all, please allow us to express our deep apologies to you. The last minor response did not satisfy you, and we have been deeply engaged in introspection on our problems. Due to the limitations on time of the modification and article size, we did not make detailed changes as you hoped. Based on your feedback that we still haven't been able to explain some parts, we have added this time. Your suggestion is absolutely right, and we should not make a subjective assumption that the reader may know something without elaborating on some terms. We have made corrections to make the article more reasonable and easier to read. In addition, for your other suggestions, we also give point-by-point responses and modification actions.
Thank you very much. Your comments have made a great contribution to the improvement of our article. While we thank you and apologize, we sincerely hope that this response will be satisfactory to you. We have also put a lot of effort into this paper and hope to get your approval. It has been very beneficial to communicate with you, and I am glad that you have guided our article.
Point 1: Most of the phrases or paragraphs mentioned (point 2) were not reformulated or given additional explanations.
Response 1: We have made relevant changes and additions. Especially for the explanation of terms and formulas.
(For details, you can see page2, line 50-54; page3, line123-127; page6, line 220-224, line246-251; page7, line 252-259, line271-292; page8, line 309-322; page9, line 327-331, line337-344; page10, line 375-377. line highlighted in yellow)
Point 2: The authors should include in the article part of the explanations given in the first review (eg points 3, 4, 5).
Response 2: Thank you for your careful reading and patient explanation, we have made relevant additions.
(For details, you can see page3, line 111-121, line 130-134, line138-144; page4, line 174-179; page13, line 470-472. line highlighted in yellow)
Point 3: The authors do not refer in the article to the measures that must be taken or that have been taken to reduce accidents, according to the mathematical study described. The authors should validate the research done with practical evidence to confirm the reduction in the number of accidents.
Response 3: We understand your question and feel sorry so much, and that's what we hope we can do in the future. However, according to China's national conditions, all measures are decided by the government, and the content needs to be taken into account to be very comprehensive, involving the economy, time, industrial restructuring and overall development, et al. For example, in a certain year, it was found that there were many deaths in coal mining enterprises, but the rapid development of solar energy made China face the suspension of coal mine production, in this case, there is no need to take any measures at all, and the death toll of coal mining enterprises will also decrease, because coal mines are gradually no longer needed in large quantities. And due to economic pressure, the government wants to do something and can't do anything.
As individual researchers, we cannot influence government decisions, so the role of proposing concrete measures is actually minimal. Our research is mainly to provide numerical value and theoretical support for the government, and the predicted numbers are to tell the decision-making organs, according to our research results, they need to make actions, but how to do it, they need to decide for themselves. The above content is actually not easy to explain in the article. That's why we can't add to your questions in great detail. However, we still think your comments very valuable, and we have added some of our predictions to the article and proved that they reflect the real situation. Although, the status quo is that we can only study theoretically possible situations, which cannot be operated in practice, your comments are still an incentive for our future research work. We will conduct in-depth research in the direction of simulating the control process and adding policy stimulus.
We have made relevant additions. It mainly complements the need to verify the validity of the forecast results of monthly, quarterly and annual construction accidents using the models we have built and to take corresponding measures.
(For details, you can see page12, line 425-433; page13, line 453-460; page14, line 483-489; page17, line 591-600. line highlighted in yellow)
Point 4: In the introduction, the authors suggest that the mathematical formulas used in the article were also used by other researchers in various fields of activity (plant accidents, traffic accidents, time change of the accident itself, road traffic accidents, production accidents, mine safety production conditions etc.), so the mathematical theoretical part should have already been verified practically. The authors should comment on the practical results and the effectiveness of the method at least through the prism of results from other fields.
Response 4: Your comments are great and we need to add them. Because many of the scholars in the introduction have already produced results in their fields using a variety of models and methods, I need to demonstrate to readers the validity of their results to demonstrate the value of the research we do. Although the mathematical formulas used in our article have been used in other fields, they are still innovative in the idea of combination model, and do not completely follow the practice of predecessors. However, it is also important to supplement the validity verification of our own research results, which helps readers grasp the focus of our article and makes the article complete and logical. Therefore, we think your comments very useful for the promotion of the article.
We have made relevant additions. (For details, you can see page2, line 54-55, line 58-61, line 70-72, line 77-80, line 85-88; page3, line 96-108. line highlighted in yellow)
Point 5: The conclusions should be completed with those described above.
Response 5: We have made relevant additions in “conclusions” section.
(For details, you can see page18, line 620-624, line 635-637, line 650-654, line 657-660. line highlighted in yellow)
Once again, thank you very much for your comments and suggestions. And we hope that the corrections will meet with approval.
Author Response File: Author Response.pdf
