Out of Alignment: Fixing Overlapping Segments in German Car Classification Through Data-Driven Clustering

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors

 

This article sets forward a new method for segmentation of vehicles by integrating technical vehicle data with clustering algorithms in order to surpass traditional, typically subjective, vehicle types such as SUV or chassis types. The goal is to develop a more precise, transparent, and adaptive classification to better guide environmental assessments and policy. But to increase the validity and transparency of your study, these careful remarks reveal some vagueness and methodological issues to be cleared up to further solidify and transfigure your manuscript.

 

1. How will you incorporate non-technical traits, e.g., ecological or policy-driven traits, into your clustering approach in future studies?
2. Did you examine whether the clusters are time-stable, in light of the fact that vehicle specifications and market shares evolve over time? With what rate would re-clustering be necessary?
3. The authors are recommended to provide more recent papers in the manuscript, such as: doi.org/10.1016/j.heliyon.2024.e33346.
4. Could the application of the specific datasets induce regional bias (Germany-specific)? To what degree are your results transferable to other markets?
5. You indicate that current vehicle classifications are largely the province of statistics rather than policy. How do you envision your clusters being used in practice in policy-making or environmental regulation?
6. In combining characteristics, how this process might disguise minute but important differences between vehicles? Would this lead to oversimplification?

 

Addressing these topics will go a long way toward increasing the contribution and reproducibility of your research.

Author Response

1. How will you incorporate non-technical traits, e.g., ecological or policy-driven traits, into your clustering approach in future studies?

   1. Thank you for your question. This is indeed a very interesting question. We currently investigate policy-changes which can have a direct influence. Regarding ecological traits, by introducing the ecological factor as one of the aggregated features, we already cover this aspect. Nevertheless, this can be expanded more towards a holistic life-cycle approach by including production and end-of-life phases. We have added statements on this in the discussion part of the manuscript.

2. Did you examine whether the clusters are time-stable, in light of the fact that vehicle specifications and market shares evolve over time? With what rate would re-clustering be necessary?
   
   1. Thank you for pointing this out! The question is exceedingly relevant, especially in the light of the two most promising methods resulting from our research (kMeans & Ward). As hierarchical clustering (Ward’s method) does not consider the market shares at all, it is somewhat stable against changing market shares. The partitioning-based method (kMeans), in turn, requires market shares to operate, so the temporal stability is reduced.
3. The authors are recommended to provide more recent papers in the manuscript, such as: doi.org/10.1016/j.heliyon.2024.e33346.

   1. Thanks for you suggestion. We critically reviewed the state of the Art one more time and added a very recent, relevant reference (https://www.mdpi.com/2071-1050/17/8/3550), while also adressing our contribution in light of the new reference. It also further supports our motivation.

4. Could the application of the specific datasets induce regional bias (Germany-specific)? To what degree are your results transferable to other markets?
   1. Thank you for addressing this aspect. We see a high transferability in the proposed methodology towards potential all other countries, especially for EU-countries, as the vehicle registration and identification is set by EU-regulations. Regarding the identified cluster, we recommend rerunning the clustering with the corresponding fleet stock data for different countries. We have added statements regarding this in the discussion part of the manuscript.
5. You indicate that current vehicle classifications are largely the province of statistics rather than policy. How do you envision your clusters being used in practice in policy-making or environmental regulation?

   1. We are in direct contact with people from the ADAC as well as the lead of vehicle statistics from the KBA. Both already showed high interest in our research, and we are happy to share this work after publication. We have added some statements regarding the usability of the work in the discussion part.

6. In combining characteristics, how this process might disguise minute but important differences between vehicles? Would this lead to oversimplification?
   

   1. Dear reviewer, thank you for addressing this issue. We are aware that by using aggregated features, we might step into an oversimplification. However, we propose both, a raw-data approach using all available data sources and an aggregated version. The aggregated approach was added to directly challenge existing segmentation logics which also utilize a subset of technical characteristics, but expanding them by more ecological features and group them interim into the proposed agg-features. This supports understandability for non-experts and potentially create more meaningful clusters, ready to use for public communication.

Reviewer 2 Report

Comments and Suggestions for Authors

This paper focused on Out of Alignment: Fixing Overlapping Segments in German Car
Classification through Data Driven Clustering. The subject matter of this manuscript aligns well with the journal’s scope, and the content appears to be original, with no indication of prior publication elsewhere. However, it is currently difficult to fully assess the contribution and impact of the study due to insufficient explanation in several key areas—namely, the significance of the research, the interpretation and practical value of the results, and the clarity of the model description.

1. Please explicitly highlight the novelty of your work. Clearly state what differentiates this study from recent state-of-the-art research and what specific advancements it offers over existing approaches.
2. The Introduction should be revised to include a more critical and comprehensive review of relevant literature. It should not only summarize prior studies but also offer critical insights into their findings and limitations. This will help justify the research gap and the motivation for your study more effectively.
3. The abstract should more clearly articulate the gap in existing segmentation methods and how the proposed machine-learning approach provides a meaningful improvement.
4. Minor language errors such as “flet” (should be fleet) and “specifiations” (should be specifications) should be corrected to maintain professionalism and readability.
5. Briefly mention the type of unsupervised learning algorithm used (e.g., k-means, hierarchical clustering) and explain what the “four engineered performance indicators” refer to.
6. While 0.19 is presented as an improvement, it would help readers to understand how meaningful this is in comparison to common clustering benchmarks.
7. Replace vague terms like “past years” with the actual time span covered (e.g., 2015–2023) to increase clarity and precision.

 

Author Response

1. Please explicitly highlight the novelty of your work. Clearly state what differentiates this study from recent state-of-the-art research and what specific advancements it offers over existing approaches.

   1. Thank you for pointing this out. We have reviewed the section and added a clearer version, directly pointing out the novelty of the proposed manuscript. Namely, by using more than one approach, we can observe differences in the cluster performance depending on the clustering algorithm. Furthermore, to investigate the performance for varying cluster sizes we are not limited to a fixed amount of clusters, which differs from the current literature. Using current date, we can incorporate the effect of the rising amount of electrified vehicles and include this in the clustering. As we propose the methodology as a framework on how to define cluster using a data-drive approach, this can be used by various others and parameters can be easily adapted.

2. The Introduction should be revised to include a more critical and comprehensive review of relevant literature. It should not only summarize prior studies but also offer critical insights into their findings and limitations. This will help justify the research gap and the motivation for your study more effectively.

   1. Thank you for your comment on this. We have added statements in the related work part and added a paragraph summarizing the limitations, thus open gaps in the current state of the art.

3. The abstract should more clearly articulate the gap in existing segmentation methods and how the proposed machine-learning approach provides a meaningful improvement.

   1. Thank you for pointing the discrepancy out. We revised the abstract, especially the explanation of the relevance of the research and the results to better reflect the specific need for that application of machine learning in this field.

4. Minor language errors such as “flet” (should be fleet) and “specifiations” (should be specifications) should be corrected to maintain professionalism and readability.

   1. Thank you for your thorough review of our manuscript. We fixed multiple small spelling errors and improved clarity and readability in certain formulations. However, we could not find the two specific examples you mentioned. We are confident that our manuscript’s language is now fine, but would be grateful if you pointed out any more deficiencies that you discover.

5. Briefly mention the type of unsupervised learning algorithm used (e.g., k-means, hierarchical clustering) and explain what the “four engineered performance indicators” refer to.

   1. Thanks for the suggestion. We elaborated further in the abstract on the methods used. Lastly we explained that the clustering is performed both on the raw technical features and separately on the engineered features. Additionally, we explained the characteristics that are captured by the four engineered features.

6. While 0.19 is presented as an improvement, it would help readers to understand how meaningful this is in comparison to common clustering benchmarks.

   1. Thanks for the suggestion. We added an additional context to clarify what these values mean and how they compare to the common clustering.

7. Replace vague terms like “past years” with the actual time span covered (e.g., 2015–2023) to increase clarity and precision.

   1. We appreciate your feedback and revised the data section accordingly.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Authors have addressed all the comments. Accept in present form.