Exploiting User Behavior to Predict Parking Availability through Machine Learning
Round 1
Reviewer 1 Report
This is definitely a good use of algorithms and machine learning to solve one of the most frustrating aspects of driving in a city environment. The use of simulation in the absence of data is innovative. And care has been taken to simulate various driver behaviors and conditions.
The machine learning results provide good predictions regarding the classification of parking availability.
From the paper the modeling takes segments of the street with statistical probability of parking availability on the segment. It may be more accurate to actually model each parking space on the street and whether it is occupied or not, and simulate drivers moving and looking for an empty space on the segment. This may allow for different anomalous conditions in the road segments such as specific events or construction, etc.
I believe more driving and parking data would make this paper much stronger.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The manuscript is related to predicting parking availability through machine learning. The study objective is interesting. The abstract can be much more informative as the methodology part is not discussed properly. The authors mentioned that "As mentioned before, our research does not consider information from the vehicle 83 speed and acceleration values, traffic conditions, or real-time data from IoT or other sensors devices." I think the authors should include all these parameters, as they are relevant in this case. The algorithm (s) are very short and not informative. The authors should share source codes, trained models, and test data for review and reproducibility checking. It will be great if they also share any web tool/software packages, etc., for review and reproducibility checking. The authors should focus on the real-time application of their algorithms. The authors used conventional approaches, it will be better if they also work on deep learning-based approaches. A comparative analysis with the existing approaches will be good. The manuscript needs minor editing in grammar and syntax. Confusion matrices have been generated on a very small number of images. It will be better if the authors include more data in the test set. Figures 11 & 12- The number of data is very small, and this result should not be considered. The authors should include more data in their study. The simulation workflow is based on rule-based approach. Why is ML/Deep learning not used?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
I am partially satisfied with the authors' responses. Some of my critical suggestions were ignored. The authors promised to work on those suggestions in their future research. My understanding is that this study is incomplete. It will be better for this manuscript if the authors consider the following points for the present study.
- Use deep learning for analysis
- Use more data for testing
- Compare your methodologies with existing approaches
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 2 Report
I read the authors' responses. The authors included texts and references and claimed that their approach is better than deep learning. The same thing they can demonstrate through their experiment. Responses to comparative analysis are unsatisfactory. The authors should include comparative results (qualitative and quantitative).
I am still not satisfied with the number of data used in figures 11 and 12. This number is not satisfactory for research purposes. There is a high chance that results will change if more data is used.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
