Exploiting Recurring Patterns to Improve Scalability of Parking Availability Prediction Systems
Round 1
Reviewer 1 Report
In this study, the authors found a high temporal auto-correlation for on-street parking dynamics in the city of San Francisco with 7-days lags. They present a processing pipeline to predict parking availability in order to reduce computational requirements. The methods used for this purpose are clustering and training set reduction techniques. I believe that the authors explain in detail the previous studies related to this topic and the paper is very well organized and easy to read. However, there are some questions I would like the authors to answer before publication and some minor mistakes. Therefore, the paper is of great interest for the journal although some minor revisions should be done.
Major comments
1. The authors used the Kennard-Stone algorithm to filter the data set. Have the authors considered using the random sampling technique instead of the Kennard-Stone algorithm to filter the data? With the algorithm selected, can the authors ensure that the data selected are uniformly distributed? Please, discuss this.
2. The authors explain in detail previous techniques, called as Reference Pipeline (RP) and state that “it is very hard for the solutions proposed in the literature to scale up to a city-side dimensions” (lines 278-286). Thus, since the authors perform two preprocessing steps in which they reduce the number of models, the questions that arise are: (a) How do the authors deal with the scaling of city-wide dimensions? (b) Are the data selected representative of the whole city?
3. According to the dendrogram, I find it confusing how the authors conclude that 5 is the number of recommended clusters. The dendrogram is cut for values above 3.5. I would like the authors to explain/clarify this in more detail. In addition, in Figures 10 and 11, the color of the clusters is different and can lead to confusion for the reader.
Specific comments
Line 2: I suggest to write “Estimated Time of Arrival” in lower case or either add the acronym “ETA”.
Line 67: I would suggest to add a reference for the Kalman filter.
Line 72: It is “Kennard” instead of “Kennerd”.
Line 75: Here the authors say 330 segments but in lines 70 and 76 there are 321 road segments. Please correct this.
Line 166: I suggest to put the reference of Zhen et al. close to the name, not at the end of the sentence.
Line 295: The equation should be revised.
Figure 5: Please, indicate the meaning of the box and the whiskers in the caption.
Table’s captions: The captions of the tables are normally above them. Please correct this for all tables.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper has interesting preliminary results for the parking availability system. The authors have said that the challenge will be to replicate the model for other city, therefore is limitation.
Although the trajectory of the work is appreciated, the results can be significantly improved by implemented other better algorithms, especially neural network or machine learning algorithms. This, I feel, should be part of the paper, rather than mentioning as future work.
What about performance comparison with other models available?
The paper length should be reduced to improve the readability of the paper, by removing the sections/contents that is not original and simply referred to.
Unnecessary self-citation should be removed.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This is a good project.
The following state-of-the-art IOT work should be mentioned in the related work
Wu, Jiaxuan, Yunfei Feng, and Peng Sun. "Sensor fusion for recognition of activities of daily living." Sensors 18.11 (2018): 4029.
Author Response
Dear Reviewer,
We would like to thank you for providing us with useful comments in this round of revisions.
We have added your suggested reference in the Related Work section, with the following sentence "Opportunistic mobile apps use smartphone sensors to estimate the subject state \cite{wu2018sensor}, [...]"
Sincerely,
Sergio Di Martino, and Antonio Origlia.
Reviewer 4 Report
- The authors analyzed the parking availability prediction service to reduce its computational requirements. They proposed a prediction model with low prediction error rates.
-The paper is well written and well organized.
-The introduction and the related work section may be merged in one section called introduction where the authors need to highlight the existing techniques and their limitations in order to convince the reader about their motivations behind this work.
-Some very old references need to be updated [1969, 2008, 2009, 2010]
- The authors may think about applying some efficient machine learning models for prediction using some public datasets, so the reader can test the proposed models as well as the results.
Overall, the paper represents great ideas with some novelty. I recommend the paper for acceptance with minor revisions.
Author Response
Dear Reviewer,
We would like to sincerely thank you for providing us with useful comments in this round of revisions.
Please find below how we addressed your comments.
In addition, we have highlighted in blue the main changes/additions made in our revised manuscript (i.e., we did not highlight minor fixes and writing issues).
Should there be any questions, please do not hesitate to bring them to our attention.
Sincerely,
Sergio Di Martino, and Antonio Origlia.
Comment: -The introduction and the related work section may be merged in one section called introduction where the authors need to highlight the existing techniques and their limitations in order to convince the reader about their motivations behind this work.
Answer: Thank you for the comment. Unfortunately we found difficulties in merging the two sections, as the resulting one was very long, affecting the readability. Nevertheless, according to your comment, in the introduction we have added new text and references to better highlight the motivations for this research.
Comment: --Some very old references need to be updated [1969, 2008, 2009, 2010]
Answer: - Some of the old references are due to seminal works in the field, like the Kennerd-Stone algorithm. We have updated older references of the State-of-art, replacing some from 2007 and 2008 with more modern papers, from 2017 and 2018.
Comment: -- The authors may think about applying some efficient machine learning models for prediction using some public datasets, so the reader can test the proposed models as well as the results.
Answer: We have highlighted in the Dataset section (3.1) that unfortunately there is no publicly available dataset with the required continuity and quality of data to compare with. We have also highlighted that we made our dataset available (to the best of our knowledge, this is the first cleansed public dataset about on-street parking availability).
Round 2
Reviewer 2 Report
The authors haven't made efforts to address the limitation pointed out during the first review round, the paper still lacks in significant technical novelty and contributions and does not show how it changes the status quo on the current parking availability prediction solutions. Only understanding the recurring data pattern on a dataset does not provide enough scientific contributions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 2 Report
I appreciate the enthusiasm of the authors. However, I am not convinced by the scientific novelty of this paper.
Upon reading the recent rebuttal letter from the authors, I read this, "we are displeased that the second reviewer is not acknowledging the scientific contribution of the paper", to clarify, my responsibility as a reviewer is not to please the authors, but to act as a critic, scrutinise the work done and provide a rigorous scientific review.
It is clear now that the contribution that authors are claiming is reducing computational complexity mostly. The author themselves have mentioned in their response that sophisticated machine learning is of no use since they do not provide improved performance but will take very long to run and increase complexity, and also authors are not interested in increasing . This is however contradictory to authors statement at the last paragraph of the paper, where authors have highlighted this as future work.
Moreover, the literature review and research gap to identify computational complexity as a problem statement is non-existent on the paper (page 2 line 38-41). The only cited work on the same problem is of the co-author themselves. This puts a big question mark on the statement of authors on the rebuttal letter, "Most of them reports severe computational issues, making it impossible to scale them to a city-wide level, and thus to deploy for the citizens".
Also, authors have neither shown any interest in comparing the work with any other work or other data sets. I do not agree that there are no available datasets. For example, the following paper does parking availability prediction on 2 datasets (SFpark and Melbourne CBD). (To note data for Melbourne is available at https://data.melbourne.vic.gov.au/)
- Zheng, S. Rajasegarar, and C. Leckie, “Parking availability prediction for sensor-enabled car parks in smart cities,” in 2015 IEEE Tenth International Conference on Intelligent Sensors, Sensor Networks and Information Processing (ISSNIP).
There is also SmartSantander project. Researchers working in this area, I would have assumed, to know about them and others.
I do not see how comparing the proposed method with state-of-the-art technique is out of scope, in fact this should be the first thing to do while claiming novelty and superiority, e.g. with the seminal paper of this field of research (has accuracy of 95%):
- Rajabioun, T., & Ioannou, P. A. (2015). On-street and off-street parking availability prediction using multivariate spatiotemporal models. IEEE Transactions on Intelligent Transportation Systems, 16(5), 2913-2924.
Not being able to produce a complete and rigorous scientific study on time for a special issue should not be the reason for paper acceptance. This is entirely based on the scientific novelty and soundness of the paper.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
