Predictive Modeling of Flight Delays at an Airport Using Machine Learning Methods
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis article presents a comprehensive study on predicting flight delays using various machine learning techniques. While the research is robust and detailed, there are several weaknesses and areas for improvement. Here are the main points:
- Dataset Limitation: The study uses data from a single Turkish airport, which may limit the generalizability of the results. Flight delay patterns can vary significantly between airports due to different operational practices, weather conditions, and traffic volumes. Expanding the dataset to include multiple airports could improve the robustness and applicability of the findings.
- Class Imbalance Handling: Although the study employs SMOTE to address class imbalance, the article does not provide a thorough evaluation of the effectiveness of this approach. It would be beneficial to compare the performance of models trained on the original imbalanced data with those trained on the balanced data created using SMOTE.
- Feature Selection: The article mentions the use of SHAP for feature selection but does not provide detailed insights into which features were deemed most important and how the feature selection process impacted model performance. A more detailed analysis and discussion of the selected features and their impact on predictions would strengthen the study.
- Model Comparison and Justification: While multiple machine learning models are compared, the rationale behind the selection of specific models (e.g., XGBoost, CatBoost, LightGBM) over others is not clearly articulated. Additionally, the discussion on why XGBoost performed the best is limited. A deeper exploration of the reasons behind the performance differences among models would add value.
- Weather Data Integration: The study integrates meteorological data but does not thoroughly examine the specific impact of different weather conditions on flight delays. A more granular analysis of how various weather factors contribute to delays could provide more actionable insights.
- Temporal Aspects: Flight delays can have temporal patterns (e.g., seasonal, daily, and hourly variations). The article does not delve into how these temporal factors were accounted for or how they influenced the predictions. Including a temporal analysis could enhance the model's predictive power and reliability.
- Economic Impact Analysis: The introduction discusses the economic impact of flight delays broadly but does not link these impacts to the findings of the study. Quantifying the potential economic benefits of the predictive model's implementation for the specific airport could illustrate the practical value of the research.
- Validation and Real-world Testing: The study lacks real-world validation or testing of the predictive models. Implementing the models in a live airport environment and reporting on their performance in real-time scenarios would provide compelling evidence of their practical utility.
- Literature Review Depth: The literature review provides a broad overview of past studies but could benefit from a more critical analysis of how the current study advances the field. Identifying specific gaps in the existing literature that this study addresses would clarify its contribution.
By addressing these weaknesses, the study could significantly improve its contribution to the field of flight delay prediction and offer more practical and generalizable insights.
Comments on the Quality of English LanguageExtensive editing of the English language is required.
Author Response
separate file is submitted
we also edit the English structure of the paper with the help of a native English speaker.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors1. The experiments were divided into four groups based on the use of SMOTE and the inclusion of weather factors. However, the impact of weather factors on the experimental results was not discussed in the previous sections, nor was it reflected in the abstract.
2. I humbly suggest avoiding the use of the first person in the abstract. Additionally, the English language needs improvement.
3. The paper primarily compares classic machine learning methods, but the title suggests the proposal of a new or improved method. I recommend modifying the title to better reflect the content.
4. The conclusion section of the abstract lacks sufficient data. It is recommended to include more data to support the conclusions.
5. Please pay attention to details, such as the missing period in line 67, and ensure consistency in writing "Naive Bayes" in lines 174 and 175.
6. Regarding the overall structure of the article, are there three issues addressed in the third part?
7. Does the title of Table 1 need to be centered, and does the third column need a title? Table 2 also has similar issues. Please follow the template requirements.
8. Both the abstract and line 60 mention the need to compare random forest algorithms, but line 201 introduces GBDT and omits random forest.
9. The second experiment used SMOTE to balance the dataset. Why does the conclusion in line 322 indicate the robustness of the three methods in handling imbalanced data?
10. I believe it is more aesthetically pleasing to have the scale line on the horizontal axis in the diagram facing upwards.
11. Four experiments have shown that XGBoost, CatBoost, and LightGBM methods are all feasible. Why, then, is XGBoost suddenly chosen for behavior analysis in line 301? Is it necessary to compare all three methods together?
12. The work presented in this paper is sufficient, but the innovation is slightly weak. It is recommended to enhance the innovative aspects.
13. The content of Part 5 is slightly redundant and needs to be condensed.
Author Response
a separate file is submitted
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper addresses the issue of flight delays, employing various machine learning methods to predict delays of over 15 minutes at a specific airport. The research is comprehensive and holds significant economic and social value.
1. Section 3 of the article, the literature review, should summarize the strengths and weaknesses of these studies and also analyze similar research (such as studies using data from only one airport).
2. There are numbering errors in the Methods section and subsequent chapters.
3. In the Methods section, in addition to describing each model, the reasons for selecting these models should be briefly explained or summarized at the end.
4. The Data set section should specify the airport or country from which the data originated and how it was obtained.
5. How many flights in the dataset were delayed?
6. After applying SMOTE technology, what is the ratio of delayed to non-delayed flight data? Is it 50% each?
7. In Tables 5 to 8, the best-performing indicators could be highlighted in bold or another color.
8. Please verify the 596 features mentioned in line 347, as the number is indeed large and seems inconsistent with what is described in the Data set section.
9. Several models in Table 8 show overfitting; please analyze this further.
10. Future research could explore the generalizability of the models, since this study uses data from a single airport. It could be useful to see if data from one airport can predict delays at other airports. Additionally, predicting the specific delay times of delayed flights could also be a valuable area of investigation.
Author Response
a separate file is submitted
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI have no additional remarks on the revised version.
The authors have addressed my concerns.
Comments on the Quality of English LanguageModerate editing of the English language is required.
Author Response
We are grateful for your contributions, which assist us in enhancing the quality of the paper. We also try to improve the English language.
Reviewer 2 Report
Comments and Suggestions for Authors1. I humbly suggest that the 73 line subheadings do not require indentation on the first line, please keep the context consistent.
2. I humbly suggest not using empty lines, such as line 208.
3. Please modify the structure of the article, as lines 136, 210, and 298 are all part three of the paper.
4. Some data in the table is bolded, what should be expressed.
5. Please note the punctuation on line 459.
6. Line 462 should be changed to Table 9.
Comments on the Quality of English LanguageThe overall English quality of this paper is good, but there are some grammatical errors and awkward phrasings. The abstract and introduction are clear, but some sentence structures are complex and should be simplified for better readability. The literature review is informative but needs more concise language and better organization. The methodology and results sections are technically sound, but some sentences are overly complex and can be simplified for clarity. Overall, it is recommended to have the paper thoroughly reviewed and edited by a professional English editor to correct grammatical errors and improve language fluency and consistency.
Author Response
- I humbly suggest that the 73 line subheadings do not require indentation on the first line, please keep the context consistent.
- thanks for the warning. We've corrected it
- I humbly suggest not using empty lines, such as line 208.
- We've removed all the unnecessary empty lines
- Please modify the structure of the article, as lines 136, 210, and 298 are all part three of the paper.
- Sorry for the numbering errors, we totally missed them.
- Some data in the table is bolded, what should be expressed.
- One of the referees suggested that the best values should be in bold. We give this information as the table footer.
- Please note the punctuation on line 459.
- We try to fix the punctuation
- Line 462 should be changed to Table 9.
- that line is based on the models with the full features (Table 5 - 8).
We also try to improve the English language.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe author has addressed the previous comments by adding content related to related to relevant research, model selection, and details of the dataset, and has also corrected some formatting issues. Overall, the author has thoroughly considered the previously raised concerns and made reasonable and comprehensive revisions. The quality of the revised manuscript has significantly improved, and the richness and validity of the content have been improved.
Author Response
We are grateful for your contributions, which assist us in enhancing the quality of the paper.
