Review Reports

Round 1

Reviewer 1 Report

-In the final paragraph of section 2.4 (lines 289-200), the goal of the study is explained by comparing it to the work of Zheng et al 2021. However, the novelty of the work in this paper and the difference between current and Zheng et all's work should be explained in a more clear manner.

-The use of COMPAS model for the AI fairness of transport modelling should be explained clearly.

-Please check the language. for instance: "For the shake of comparability,..."

-Line 277, please check the equation.

-Please correct the references. For instance, Zheng, Y., Wang, S., & Zhao, J. (2021) is a very important paper in this field and it has already been published in a journal.

Author Response

Response to reviewer 1 comments

Point 1: In the final paragraph of section 2.4 (lines 289-200), the goal of the study is explained by comparing it to the work of Zheng et al., 2021. However, the novelty of the work in this paper and the difference between current and Zheng et al., 2021 work should be explained in a more clear manner.

Response 1: This comment is highly appreciated as it is key to make clear what the contributions of this to the existing literature is. We have included an additional paragraph (Section 2.4, lines 277 – 287), making a more explicit description of how our work differs from the one of Zheng et al, 2021 and what are the specific novelties of the study we have conducted. We have made special emphasis on the utilization of three different bias mitigations algorithms that have been implemented in three different stages of the model development process. Furthermore, we believe that the inclusion of both survey data and opportunistic data is a valuable novelty, as it allows to compare how the impact of bias might be dependent on the kind of source from which the training data has been obtained.

Point 2: The use of COMPAS model for the AI fairness of transport modelling should be explained clearly.

Response 2: Additional explanations have been included (Section 3.1, lines 297 – 301) clarifying that the COMPAS model has been analysed in order to confer a certain level of comparability with the previous literature to our results, as this model has already been studied in several works dealing with the topic of algorithmic fairness.

Point 3: Please check the language. for instance: "For the shake of comparability,..."

Response 3: The whole paper has been reviewed again to modify any expression that could be unclear or incorrect.

Point 4: Line 277, please check the equation.

Response 4: the equation has been corrected.

Point 5: Please correct the references. For instance, Zheng, Y., Wang, S., & Zhao, J. (2021) is a very important paper in this field, and it has already been published in a journal.

Response 5: We have included the full reference to the paper, including the journal it has been published in. 

Reviewer 2 Report

1) Your study lacks a clearly defined goal and a hypothesis (s) or research question (s).

2) What is the practical application of your research? Who can use your work in practice? Will the practitioners working in the city be able to apply your results?

Author Response

Response to Reviewer 2 Comments

Point 1: Your study lacks a clearly defined goal and a hypothesis (s) or research question (s).

Response 1: We have incorporated several considerations into the paper aiming to further clarify which are research questions and the goal of our work. The final part of the introduction (Section 1, lines 95 – 104) has been modified in order to explicitly formulate which are the research questions that our work aims to answer. First, we would like to determine which of the existing techniques maximizes the mitigation of bias while minimizing the loss of classification accuracy. Second, we would like to determine if the use of training data that has been acquired through uncontrolled methods entails higher levels of algorithmic bias on mode choice models, than survey data.

Point 2: What is the practical application of your research? Who can use your work in practice? Will the practitioners working in the city be able to apply your results?

Response 2: We would like to thank the reviewer for this comment. We believe that the results of our research are mainly useful for the community of transport modellers, rather than for the practitioners working in the city authorities or policy makers. Our goal is to provide first results on what is the level of algorithmic bias that might affect two sample mode choice models that could potentially be used for transport policy support and how effective the existing bias mitigation techniques are, when implemented in the modelling pipeline. Therefore, our research intends to be a relevant contribution to the ethical and accurate implementation of machine learning techniques on transport modelling, more than to provide policy recommendation that can be directly applicable.

To better clarify this point, further justification of the relevance of our work and our results has been included in the manuscript (Section 6, lines 769 – 837).

Reviewer 3 Report

Page 2, Keywords: My suggestion is to include “behavioural transport models” in your keywords.

Table 3: Sample distribution of the survey data used in the Active Modes model: Please note that the sum of the percentages in the case of “Education” is 100.1% and not 100%, the sum of the percentages in the case of “Size of the city” is 99.9% and not 100% and finally, the sum of the percentages in the case of “Area of residence” is 99.9% and not 100%. Please correct the specific percentages.

Table 4: Sample distribution of the multimodal trip planner data used in the Beijing model – trip characteristics: Please note that the sum of the percentages in the case of “Day of the week” is 98.6% and not 100% and the sum of the percentages in the case of “Hour of the day” is 99.9% and not 100%. Please correct the specific percentages.

 It seems that the fonts used in Section 5. Discussion (last 3 paragraphs with bullet points) are different from the rest.

Author Response

Response to Reviewer 3 Comments

Point 1: Page 2, Keywords: My suggestion is to include “behavioural transport models” in your keywords.

Response 1: We appreciate this suggestion, given than this works focus on the transport modelling from the point of view of the individual’s behaviour. Therefore, we have substituted the term “transport modelling” for the suggested one (“behavioural transport models”), which is more precise. 

Point 2 and 3:

Table 3: Sample distribution of the survey data used in the Active Modes model: Please note that the sum of the percentages in the case of “Education” is 100.1% and not 100%, the sum of the percentages in the case of “Size of the city” is 99.9% and not 100% and finally, the sum of the percentages in the case of “Area of residence” is 99.9% and not 100%. Please correct the specific percentages.

Table 4: Sample distribution of the multimodal trip planner data used in the Beijing model – trip characteristics: Please note that the sum of the percentages in the case of “Day of the week” is 98.6% and not 100% and the sum of the percentages in the case of “Hour of the day” is 99.9% and not 100%. Please correct the specific percentages.

Response 2 and 3: We would like to thank the reviewers for noticing these mistakes. We have already checked the percentages for all the sample distribution tables to ensure that they are correct.