Short-Term Traffic-Flow Forecasting Based on an Integrated Model Combining Bagging and Stacking Considering Weight Coefficient
Round 1
Reviewer 1 Report
The manuscript titled “Short-Term Traffic Flow Forecasting Based on the DW-Ba-Stacking Model” treats several approaches to the short-term traffic flow forecasting problem and proposes a dynamic weighting adjustment model, coupled with stacking and bagging.
The presentation is fluid and well-argued. Some additional references to the literature would be required for supporting arguments (to be seen below).
Remarks:
- From Section 2.1.3.1 (Line 117), there is a jump to Section 3.1.3.2 at Line 147.
- Section 3.1.3.3 – Line 189: “Regular sequences have won the favour of various scholars.” Perhaps a reference to the literature would be appropriate here.
Similarly on Line 192: “Then some scholars proposed the LSTM model can effectively solve this problem, but there are too many parameters to adjust in the network training.”
- In line 213, there is a return to Chapter 2.2. Please revise the chapter, sections numbering.
- Line 214: “The overall architecture of the DW-Ba-Stacking model consists of a Bagging model processing phase and a Stacking model processing phase, with the Stacking model architecture playing a large role as the Bagging model is only embedded in the Stacking model.” Perhaps reformulate for a better clarity.
- Line 255: “The addition of 0.5 to the Ln function in Equation 10 is to accommodate the calculation of zeros in the original series.” The need for a positive argument in the case of the logarithmic function is clear; it would be useful to justify the specific constant 0.5.
- Formulas (13) and (15) would need some additional spacings between terms for better clarity.
- Line 330: “Bagging algorithm, denoted as 1X Y 、2 X Y 、3X Y 、4 X Y 、5 X Y and 6 X Y ;”. Please, make sure the terms are separated with comas. Perhaps create an equation form if necessary. Similarly, at lines 331, 333, and 334.
- Line 392: “To explore the temporal characteristics of traffic flow in depth, the trend of traffic flow changes over a period of time is randomly selected for analysis, as shown in Figure 8.” The meaning to be conveyed is not clear.
- Line 475: Need for introduction of the abbreviation like: Xgboost, GBDT etc.
- Figure 13 and Table 6 refer to the “Entropy method combination”, but this concept does not appear to be developed in the manuscript.
Suggestions:
- I suggest that the DW-Ba-Stacking model be clarifying the meaning designation earlier in the presentation. The authors go thoroughly through it in Chapter 2, but are already referring to it by then.
- The reference to the literature needs to be broader in scope and to go into some comparative details.
- Generally, the manuscript has a value that needs to be better emphasized. The difference between the investigated versions of the forecasting algorithms to be presented in a comparison with industry standards.
Author Response
Point 1: From Section 2.1.3.1 (Line 117), there is a jump to Section 3.1.3.2 at Line 147.
Response 1: We have checked the chapter number and have revised it.
Point 2: Section 3.1.3.3 – Line 189: “Regular sequences have won the favour of various scholars.” Perhaps a reference to the literature would be appropriate here.
Similarly on Line 192: “Then some scholars proposed the LSTM model can effectively solve this problem, but there are too many parameters to adjust in the network training.”
Response 2: We rephrased this section in section 3.1.3.3 and added relevant references.
Point 3: In line 213, there is a return to Chapter 2.2. Please revise the chapter, sections numbering.
Response 3: We have checked the chapter number and have revised it.
Point 4: Line 214: “The overall architecture of the DW-Ba-Stacking model consists of a Bagging model processing phase and a Stacking model processing phase, with the Stacking model architecture playing a large role as the Bagging model is only embedded in the Stacking model.” Perhaps reformulate for a better clarity.
Response 4: We were not clear at the beginning, and have revised it to:" The overall architecture of the Ba-Stacking model included the Bagging model processing stage and the Stacking model processing stage. Because the bagging was only embedded as part of the stacking model, the stacking model architecture plays a big role".
Point 5: Line 255: “The addition of 0.5 to the Ln function in Equation 10 is to accommodate the calculation of zeros in the original series.” The need for a positive argument in the case of the logarithmic function is clear; it would be useful to justify the specific constant 0.5.
Response 5: The setting of 0.5 is to make the data show a more uniform trend and avoid the possibility of affecting the prediction accuracy due to uneven distribution, which is at Line 255 in original article. And no scholar has studied it specifically, this is an experience value.
Point 6: Formulas (13) and (15) would need some additional spacings between terms for better clarity.
Response 6: We have rewritten the Formula(13) and (15) to make them clearer.
Point 7: Line 330: “Bagging algorithm, denoted as 1X Y 、2 X Y 、3X Y 、4 X Y 、5 X Y and 6 X Y ;”. Please, make sure the terms are separated with comas. Perhaps create an equation form if necessary. Similarly, at lines 331, 333, and 334.
Response 7: We have rewritten the formula to make it clearer, which is at Lines 330,331,333 and 334 in original article.
Point 8: Line 392: “To explore the temporal characteristics of traffic flow in depth, the trend of traffic flow changes over a period of time is randomly selected for analysis, as shown in Figure 8.” The meaning to be conveyed is not clear.
Response 8: The translation of the original text is inaccurate and has been corrected as "To explore the temporal characteristics of traffic flow in depth, the trend of traffic flow changes over a period of time is randomly selected for analysis. The selected data is shown as shown in Figure 8", which is at Line 392 in original article.
Point 9: Line 475: Need for introduction of the abbreviation like: Xgboost, GBDT etc.
Response 9: The original text has introduced abbreviations, changed to "In order to analyze the effects of historical related characteristics, some single models such as XGBoost ,GBDT, etc. are selected for comparative analysis, shown as Table 1", which is at Line 475 in original article.
Point 10: Figure 13 and Table 6 refer to the “Entropy method combination”, but this concept does not appear to be developed in the manuscript.
Response 10: Ihe chart sorted out some problems at beginning, Figure 13 and Table 6 have deleted "Entropy method combination".
Suggestions:
Suggestion 1: I suggest that the DW-Ba-Stacking model be clarifying the meaning designation earlier in the presentation. The authors go thoroughly through it in Chapter 2, but are already referring to it by then.
Response 1: We changed the way the model was described before Chapter 2 and used the DW-Ba-Stacking model after a thorough introduction in Chapter 2.
Suggestion 2: The reference to the literature needs to be broader in scope and to go into some comparative details.
Response 2: We have added more references in Chapter 1 and introduced some comparative details
Suggestion 3: Generally, the manuscript has a value that needs to be better emphasized. The difference between the investigated versions of the forecasting algorithms to be presented in a comparison with industry standards.
Response 3: We compared with some methods in other references and designed comparative experiments, which have been described in section 4.2.6
Reviewer 2 Report
This paper DW-Ba-Stacking model for short-time traffic flow. This paper is well organized, and the topic is interesting. The following comments should be further considered to improve the paper:
- Do not use the unwell-known abbreviations, DW-Ba, in the title and abstract.
- This paper mentions LSTM prediction purposes. The advantage of LSTM should be mentioned by reviewing the deep learning-related work, e.g., A Transfer Learning-based State of Charge Estimation for Lithium-Ion Battery at Varying Ambient Temperatures.
- The introduction makes a literature review of machine learning-based predictions. More related work, e.g., Slow-varying Dynamics Assisted Temporal Capsule Network for Machinery Remaining Useful Life Estimation, can be added to make the review comprehensive.
- The authors are suggested to carefully proof the whole manuscript to avoid grammar errors and typos.
Author Response
Point 1: Do not use the unwell-known abbreviations, DW-Ba, in the title and abstract.
Response 1: We have changed the title and abstract to make it easier to understand.
Point 2: This paper mentions LSTM prediction purposes. The advantage of LSTM should be mentioned by reviewing the deep learning-related work, e.g., A Transfer Learning-based State of Charge Estimation for Lithium-Ion Battery at Varying Ambient Temperatures.
Response 2: We added a review of deep learning related literature such as " Transfer Learning-Based State of Charge Estimation for Lithium-Ion Battery at Varying Ambient Temperatures." in reference 10.
Point 3: The introduction makes a literature review of machine learning-based predictions. More related work, e.g., Slow-varying Dynamics Assisted Temporal Capsule Network for Machinery Remaining Useful Life Estimation, can be added to make the review comprehensive.
Response 3: We added a review of deep learning related literature such as "Slow-varying Dynamics Assisted Temporal Capsule Network for Machinery Remaining Useful Life Estimation." in reference 15.
Point 4: The authors are suggested to carefully proof the whole manuscript to avoid grammar errors and typos.
Response 4: We have re-proofed the full text to correct grammatical errors and typos.
Reviewer 3 Report
In this paper, a DW-Ba-Stacking model for short-time traffic flow prediction is presented. From my view, this paper is not well organized and the proposed method is not valuable for this research filed. After reviewed this paper, there are some questions and suggestions as follows.
1- The literature review is poor in this paper. You must review all significant similar works that have been done. Also, review some of the good recent works that have been done in this area and are more similar to your paper.
2- It is necessary to talk about the role of the parameters of the proposed algorithm in a separate section. For example, which parameters are responsible for controlling diversity of ensembeling?
3- In the part of results and discussion, the author should compare the recent relevant work in ensemble learning field with the results of this paper to confirm the effectiveness of this study.
4- It is necessary to experimentally analyze the proposed algorithm in terms of time consumed and compare with other algorithms.
5- What are the advantages and disadvantages of this study compared to the existing studies in this area?
6- There are many grammatical mistakes and typo errors.
7- In Figure 7, are the training and test data used for model training at the same time?
8- Formula 19 is wrong.
9- Write a pseudocode for the proposed algorithm.
Author Response
Point 1: The literature review is poor in this paper. You must review all significant similar works that have been done. Also, review some of the good recent works that have been done in this area and are more similar to your paper.
Response 1: We added some reviews of recent related literature in Section 1.
Point 2: It is necessary to talk about the role of the parameters of the proposed algorithm in a separate section. For example, which parameters are responsible for controlling diversity of ensembeling?
Response 2: This work does not involve too many parameters.
Point 3: In the part of results and discussion, the author should compare the recent relevant work in ensemble learning field with the results of this paper to confirm the effectiveness of this study.
Response 3: We compared with some methods in other references and designed comparative experiments, which have added some descriptions in section 4.2.6.
Point 4: It is necessary to experimentally analyze the proposed algorithm in terms of time consumed and compare with other algorithms.
Response 4: Our model does not account for time-dependent studies, which is where we need further improvement and has been formulated in Section5.
Point 5: What are the advantages and disadvantages of this study compared to the existing studies in this area?
Response 5: We have carefully reviewed our work and analyzed the strengths and weaknesses of this paper in Section 5.
Point 6: There are many grammatical mistakes and typo errors.
Response 6: We have re-proofed the full text to correct grammatical errors and typos.
Point 7: In Figure 7, are the training and test data used for model training at the same time?
Response 7: The first draft did not express clearly, and added the expression " Training the model with the training set. Once trained, the model is tested using the test set " in section 2.4.2.
Point 8: Formula 19 is wrong.
Response 8: We have rewritten the formula to make it right.
Point 9: Write a pseudocode for the proposed algorithm.
Response 9: The optimization process has been described in section 2.4.2 and demonstrated in Figure 7.
Round 2
Reviewer 3 Report
Good revisions have been made in the paper and the revised version has the necessary qualities for acceptance compared to the previous version. In my opinion, the article is acceptable in its current form.
