Spatial Models and Neural Network for Identifying Sustainable Transportation Projects with Study Case in Querétaro, an Intermediate Mexican City

Round 1

Reviewer 1 Report

The topic taken into account in the paper is interesting and worth publication. In the reviewed paper, the Authors presented the spatial models and neural networks for estimating public transport demand in urban avenues. In the reviewed paper, the Authors presented a comparative analysis of spatial models using artificial intelligence to estimate transport demand. For doing that, they present empirical research in certain urban corridors to measure the selected indicators. A study case is located in Querétaro, an emergent city of the Mexican region known as El Bajío. Two near and similar urban avenues in width and length are selected to apply a group of spatial models, evaluating the avenues by segments and predicting public transport demand. The resulted database is analyzed by the Authors by using Artificial Neural Networks, and it displays that specific indicators have around 80% of correlation. The results facilitate the localization of the avenue segments with the most volume of activity, supporting interventions in urban and transport renewal projects. In my opinion, the paper can be published, after taking into account the following remarks:

• before paper printing, English should be carrefully checked by a Native Speaker,
• in the keywords section, the keyword "sustainable transportation" should be added,
• the paper is dedicated to public transport demand prediction. In the present Introduction section, the Authors described the need for conducting urban analyzes, spatial models for road infrastructure, stated the problem of which spatial model better correlates to the actual public transport demand, etc. Unfortunately, the Authors did not address the context of the problem by describing the elements that contribute to the relevant public transport demand in the cities like park and ride solutions and bike-sharing systems. Authors should mention these popular solutions and refer to the latest scientific literature on the subject in this regard, i.e. "The analysis of the factors determining the choice of park and ride facility using a multinomial logit model", doi 10.3390 / en14010203; "A P-Hub Location Problem for Determining Park-and-Ride Facility Locations with the Weibit-Based Choice Model", doi.org/10.3390/su13147928; P&R parking and bike-sharing system as solutions supporting transport accessibility of the city, doi10.21307 / tp-2020-066p; One short paragraph in the paper will be enough (in the Introduction section),
• Figure 1. PSAT indicator on selected urban avenues - the legend is written in too small letters, so is invisible. It should be improved. The same remark is dedicated to figure 2, and figure 5,
• subsection called "3.2. Indicator of Spatial and Functional Continuity of Street Corridor" should not be divided into further sub-sub sections (3.2.1, and 3.2.2) because their content is too short. The Authors could either develop the content of each sub-sub section or just not divided subsection 3.2 into further sub-sub sections,
• figure 3, figure 4 - the legend is not in English,
• the paper text is not formatted according to the Sustainability journal requirements, e.g. tables,
• line 294: the Authors wrote like follows: ..."In this exercise"... This is a paper, not exercise,
• there is a lack of solid discussion dedicated to obtained research results. It should be improved.

Author Response

Reviewer 1

 

We sincerely appreciate your taking the time to review our work. We greatly appreciate each of your comments that help improve the quality of our work. Considering your comments, we have made the following responses:

 

1. In the keywords section, the keyword "sustainable transportation" should be added,

Thank you for the observation. The keyword was added.

2. The Authors did not address the context of the problem by describing the elements that contribute to the relevant public transport demand in the cities like park and ride solutions and bike-sharing systems. Authors should mention these popular solutions and refer to the latest scientific literature on the subject in this regard, i.e. "The analysis of the factors determining the choice of park and ride facility using a multinomial logit model", doi 10.3390 / en14010203; "A P-Hub Location Problem for Determining Park-and-Ride Facility Locations with the Weibit-Based Choice Model", doi.org/10.3390/su13147928; P&R parking and bike-sharing system as solutions supporting transport accessibility of the city, doi10.21307 / tp-2020-066p; One short paragraph in the paper will be enough (in the Introduction section),

We appreciate this wise observation. We added a paragraph in the introduction section line 47-49, and it supports explaining solutions for the public transportation demand.

3. Figure 1. PSAT indicator on selected urban avenues - the legend is written in too small letters, so is invisible. It should be improved. The same remark is dedicated to figure 2, and figure 5,

We appreciate this important recommendation. We made the changes for all figures.

4. subsection called "3.2. Indicator of Spatial and Functional Continuity of Street Corridor" should not be divided into further sub-sub sections (3.2.1, and 3.2.2) because their content is too short. The Authors could either develop the content of each sub-sub section or just not divided subsection 3.2 into further sub-sub sections,

We appreciate your valuable comment. We deleted the sub-sub-sections and put the titles inside the paragraph so it makes more sense for the reading. This in lines 159, and 179. 

5. figure 3, figure 4 - the legend is not in English,

We are grateful for your comment. We made the changes for the figures, in line 118 and 130.

6. the paper text is not formatted according to the Sustainability journal requirements, e.g. tables,

We appreciate this important recommendation. We made the change to the formatted text and all the tables.

7. line 294: the Authors wrote like follows: ..."In this exercise"... This is a paper, not exercise,

We appreciate this wise observation. We corrected the writing and now it is in line 346.

8. There is a lack of solid discussion dedicated to obtaining research results. It should be improved.

We are grateful for your comment. We added to the discussion in lines 320 - 329, that models with few variables and simpler complexity, like the Range of Activities which uses one variable, have similar results to more complex models like the PSAT tool, which uses 14 variables. We added the importance to noting that the PSAT tool has 1.92 % more assertion in its prediction, and its complexity of having 14 variables means that it can function as a classifier. This means it can specify which kind of activities are in the territory, rather than just locate the most, as the Range of Activities does.

In lines 336-337, another thing to comment, is the importance of locating the projects from local streets rather than an urban avenue, alluding to the last-mile type of urban renewal projects. 

Then, in lines 341-344, we added that this kind of projects support sustainable transport as well as other solutions like the park and ride or the bike share systems.

Reviewer 2 Report

In the section entitled: 4. Method: Phase B. Comparative analysis using neural network it is necessary to state which software to work with Artificial Neural Network (ANN) was used to test selected architectures. Also, I suggest that the measure of performance Root-Mean-Square Error (RMSE) be precisely formulated by a mathematical formula. It would be useful to compare the time required to solve the problem using spatial models and using neural networks.

Author Response

Reviewer 2

We appreciate that you have shared your valuable time and knowledge to review our work. Your respected recommendations have been studied extensively and the responses we have made are shown below.

In the section entitled: 4. Method: Phase B. Comparative analysis using neural network it is necessary to state which software to work with Artificial Neural Network (ANN) was used to test selected architectures. Also, I suggest that the measure of performance Root-Mean-Square Error (RMSE) be precisely formulated by a mathematical formula. It would be useful to compare the time required to solve the problem using spatial models and using neural networks.

We are grateful for your comment. The software was added in line 256. The RMSE equation was added in line 281. In this case, the network is used to infer the degree of prediction of the models. The processing time of the models for each point is negligible. The only considerable computation time will be that of network training. This is explained to the reader online 256-260.

Reviewer 3 Report

In this manuscript, the authors just make a comparative analysis of spatial models using artificial intelligence to estimate transport demand. There is no innovation idea found or a clear view of what the authors have done as a contribution. Therefore, I don’t think this paper can be accepted in its current form.

 

 

1. This manuscript simply presents a comparative analysis of four models that have been constructed to estimate the public transport demand in urban avenues. The work is easy and not innovative.

 

2. It is recommended to write out the formulas of the three transfer functions. Besides, what is the significance of choosing the three specific functions? Are other functions also effective in evaluating the performance of the models?

 

3. Wouldn’t it be better if the tables appearing in the manuscript were three-line-tables? In addition, the authors are encouraged to compare the running time of all models or add analysis of the computational complexity.

 

4. The conclusion and the future work are missing at the end of this paper, it should be supplied to further improved the quality of this article.

 

5. English sentence construction is poor, and language proofreading is definitely required. All mistakes, such as the one that appears in line 39, are avoided in the paper as much as possible.

Author Response

Reviewer 3

We are deeply grateful for the opportunity to have our work reviewed by you. We appreciate each of your meaningful comments sharing your knowledge. We have tried to respond to your comments below.

1. There is no innovation idea found or a clear view of what the authors have done as a contribution.

We appreciate this wise observation. We added in lines 126-127, a reference that the municipal government wants to intervene in one of the urban avenues of the paper, specifically Ave. Pirineos. This is the one with lesser values in all of our indicators. These results help to orient and change the strategy of urban renewal projects. 

Also, we added a final map of the area, line 312-316, with the results of the PSAT indicator, like the one that was implemented in all the metropolitan street network. Here it can clearly see where to do intervention projects. For instance, it is needed around the main corridor that contributes to feeding the public transport system, and not in Ave. Pirineos. In lines 336-337, we added the importance of locating the projects from local streets rather than an urban avenue, alluding to the last-mile type of urban renewal projects.

2. This manuscript simply presents a comparative analysis of four models that have been constructed to estimate the public transport demand in urban avenues. The work is easy and not innovative.

We appreciate this wise observation. In lines 42-45, we added an explanation of the objective of every model, describing that they are not constructed for estimation of public transport, but for general qualification of an urban corridor. That is why we decide to compare them with a natural activity of an urban corridor, such as the public transport demand.

Also, we added in the discussion section, lines 337-340, that one of the contributions is the location of urban renewal projects for an entire metropolitan area. The data available and the cost of the process made possible the results, instead of using extensive and expensive field works. With the neural network approach, there is not research reported in the state of the art which has a simmilar.

3. It is recommended to write out the formulas of the three transfer functions. Besides, what is the significance of choosing the three specific functions? Are other functions also effective in evaluating the performance of the models?

We are grateful for your comment. Formulas for the three transfer functions have been added, and the reason for their choice is occasionally explained on lines 281- 283.

4. Wouldn’t it be better if the tables appearing in the manuscript were three-line-tables? In addition, the authors are encouraged to compare the running time of all models or add analysis of the computational complexity.

We appreciate your valuable comment. The tables have been modified. The urban indicators are calculated with algebraic models, so the computation time is negligible, as is the execution time of the trained network. On the other hand, the network training times can inform computation times depending on the number of data and the architecture used. This is made clear by reading lines 258-259.  

5. The conclusion and the future work are missing at the end of this paper, it should be supplied to further improve the quality of this article.

We appreciate this important recommendation. We added extensive comments in lines 345-371. We made a summary of the paper, enlisted the principal results of the indicators and the correlations, like: the contributions about proper location of urban renewal projects, related to volume of people using these locations actually, all this using data available and a low cost technique instead of extensive and expensive resources; and finally the future work in lines 371-378, about extending the study to other cities.

6. English sentence construction is poor, and language proofreading is definitely required. All mistakes, such as the one that appears in line 39, are avoided in the paper as much as possible.

We appreciate your valuable comment. We reviewed the manuscript with an English native professional.

Reviewer 4 Report

Studies regarding public transport demand in urban areas are necessary for a better and well designed development of the cities. In some parts of the World, where cities development are faster, these studies must be made in order to adjust the transport infrastructure to city extension. 

The article is good, but need some corrections, especially to make data used more reasonable. In some parts of the paper is difficult to understand from where are come the data. Table 1 to what of the considered avenue makes reference?

Also, avenues configurations are different and a comparative analyse can be not relevant for the study purpose. Maybe, a better analyse must be made using avenues with closer characteristics.

English language and style need some minor corrections.

Author Response

Reviewer 4

We sincerely appreciate your taking the time to review our work. We greatly appreciate each of your comments that help improve the quality of our work. Considering your comments, we have made the following responses:

 

1. The article is good, but need some corrections, especially to make data used more reasonable. In some parts of the paper is difficult to understand from where are come the data.

We are grateful for your comment. We added more explanations on the data used for the models; The majority of the information is available by the INEGI, Instituto Nacional de Estadística y Geografía, in line 143 for the Indicator Public Space Distribution; in lines 166-168 for the Indicator of Range of Activities; and in line 233-234 for the Indicator of balance between residence and activities; Finally we made the field works in Ave. de la Luz and Ave. Pirineos for the public transport demand data, in lines 240-245.

2. Table 1 to what of the considered avenue makes reference?

We appreciate your valuable comment. It is for Av. de la Luz, and also we added a table for Av. Pirineos, and an explanation in lines 197-201.

3. Also, avenues configurations are different and a comparative analyse can be not relevant for the study purpose. Maybe, a better analyse must be made using avenues with closer characteristics.

We appreciate your valuable comment. We added an explanation about the selection of both avenues, in lines 112-114. 

Apart from that, currently there is an urban renewal project of the public administration that is in Ave. Pirineos in lines 126-127.

Also, in lines 312-316, we added a final map of the area, with the results of the PSAT indicator, as the one that was implemented in all the metropolitan street network. Here it can clearly see where to do intervention projects. For instance, it is needed around the main corridor that contributes to feeding the public transport system, and not in Ave. Pirineos. The results of the paper support a change of strategy for this and other projects.

4. English language and style need some minor corrections.

We are grateful for your comment. We reviewed the manuscript with an English native professional.

Round 2

Reviewer 3 Report

Although the authors have revised the manuscript according to the review comments,  none of the substantive content, such as the innovative points of this paper, has changed.

1. “Using artificial neural networks (ANN) to analyze the collected date and obtaining the fact that there is a high correlation between certain indicators” is the core of this paper? If so, it would be too simplistic
2. It can be seen that the three transfer functions are illustrated. Please elaborate on the specific meaning of the parameters in these transfer functions. In addition, wouldn’t it be better to place the figures and tables in the middle of pages?
3. The structures of the ANN utilized in the paper are small, would deeper ANN give a more accurate prediction of public transport demand in urban avenues?
4. The authors only used one metric, namely RMSE, to evaluate the performance of models. Please compare the running time of all models or add analysis of the computational complexity, too.
5. The experimental section is too empty and prone to prevent other researchers from reproducing the results and conducting further research in related areas. It is hoped that the experimental configuration, such as the experimental platform, can be described in detail.

Author Response

Reviewer 3

We sincerely appreciate your taking the time to review our work. We greatly appreciate each of your comments that help improve the quality of our work. Considering your comments, we have made the following responses:

1. “Using artificial neural networks (ANN) to analyze the collected data and obtaining the fact that there is a high correlation between certain indicators” is the core of this paper? If so, it would be too simplistic

Thank you for the observation. The main core and contribution of the paper are about identifying the street segments with the potential of people using them in a sustainable way, and therefore, for government or private interventions. This paper is comparing a zone of two urban avenues and also comparing certain indicators using artificial intelligence for the latter objective. The results determine that there is a need for urban projects around one of the avenues analyzed. As one of the references included in lines 126-127, the government is instead of spending on an intervention project in the avenue with no need. 

We also added in lines 141-147, the variables of the indicator that has the best correlation, in order to give more clarity to the process. These variables put qualifications in every street segment of the entire road network in the city. The variables are road incidence, infrastructure for the disabled, diversity of transport modes, block length, street network grid, and mixed land use. Other variables were considered as proximity to housing, jobs, schools, commerce, presence of trees, urban lights, road signals, and sidewalks. The amount of variables used in this indicator implies a more complete picture of the work made.

Finally, there is no reported research using both indicators and artificial intelligence in this short-scale approach.

2. It can be seen that the three transfer functions are illustrated. Please elaborate on the specific meaning of the parameters in these transfer functions. In addition, wouldn’t it be better to place the figures and tables in the middle of pages?

Thank you for the comment.  Figure 6 was added better to explain the parameters of the functions and their use. Additionally, the parts of a neuron, the role of the transfer function, and the input and output parameters are described. The changes are observed in lines 278-283.

3. The structures of the ANN utilized in the paper are small, would deeper ANN give a more accurate prediction of public transport demand in urban avenues?

We appreciate the comment. The focus of this research is based on Machine Learning, and Deep Learning has been left aside. The main reason is that, as is well known, a greater amount of data is required for a deep approach. That is explained to the reader in line 283-287. It is also a reference to a paper that has extensive data of the public transport demand, provided by the official government, and the results effectively have a more accurate prediction. This paper takes a shorter scale approach, with lesser public transport data, because the cost of taking is high in time and resources. This is explained to the reader in lines 149-153. 

4. The authors only used one metric, namely RMSE, to evaluate the performance of models. Please compare the running time of all models or add analysis of the computational complexity, too.

We appreciate your valuable comment. Although the MSE and the RMSE are usually the main performance indicators in the vast majority of articles related to machine learning, we add the R2 and the MBE to address the suggestion. The R2 represents the quality of the model to predict the results, and the MBE the bias in the model. These two statistical indicators, together with the RMSE, provide the reader with a complete idea of the predictive value of the ANN. The computation time was not considered since it depends on the computer where it is executed. Also, the algorithm is implemented in a general-purpose computer, which implies significant changes in the process according to the current load of the secondary processes. On the other hand, it is well known that the higher the number of neurons and the number of layers, the higher the computational cost. The changes are observed in lines 313-315.

5. The experimental section is too empty and prone to prevent other researchers from reproducing the results and conducting further research in related areas. It is hoped that the experimental configuration, such as the experimental platform, can be described in detail.

We are grateful for your comment. We added the variables of the indicator that has the best correlation, in order to permit the reproduction of the methodology in other cities, in lines 141-147. We also added a flow chart in Fig. 1, to better describe the general process. Finally, we added a reference in lines 149-153, to a bigger exercise in a metropolitan area, using the same indicator, with better results.

Round 3

Reviewer 3 Report

Although the authors have made further changes to the manuscript according to the review comments, the substantive content, such as the innovative points of this paper, remain unchanged. Therefore, I think the authors can seek other journals for submission rather than continuing to revise the article.

Author Response

Thank you for your comments. I am attaching the comments made by the academic editor which contain your concerns:

We sincerely appreciate your taking the time to review our work. We greatly appreciate each of your comments that help improve the quality of our work. Considering your comments, we have made the following responses:

 

1. Studies regarding public transport demand in urban areas are necessary for a better and well designed development of the cities. In some parts of the World, where cities develop faster, these studies must be made in order to adjust the transport infrastructure to city extension. 

Thank you for the observation. We added in lines 399-401 of the conclusions, a reference of the size of the city, along with the fact that it is in a developing country, implies that technology is obsolete to accurately locate the volume of people moving in and out of a public transport system. 

2. The article is good, but needs some corrections, especially to make the data used more reasonable. In some parts of the paper, it is difficult to understand where the data come from. Table 1 to what of the considered avenue makes reference?

Thank you for the comment. We added a reference for the avenue in Table 1 and Table 2.

3. Also, avenues configurations are different and comparative analysis can be not relevant for the study purpose. Maybe, a better analysis must be made using avenues with closer characteristics

We appreciate the comment. In lines 332-345 we added two avenues with closer characteristics, and we compare them with the indicator that has the better results of the paper. The results display a similar need for intervention in these urban corridors, as well as in locations around the avenues.

4. English language and style need some minor corrections.

We appreciate your valuable comment. We revised the language with a native expert.

5. This is a case study, which should be reflected in the title. This is a study case in Querétaro, an emergent city in the Mexican region known as El Bajío

Thank you for the comment. We adjust the title as a study case and the city.

6. The authors should also address the study limitations clearly.

We appreciate your valuable comment. We added in lines 406-408 the r2 results as part of the limitations that need to be addressed in subsequent investigations.