Spatiotemporal Prediction of Theft Risk with Deep Inception-Residual Networks
Round 1
Reviewer 1 Report
In the revised version of the manuscript, the authors took into account all my comments.
One minor comment:
1. In lines 407-411 the formatting of the text should be corrected.
Very interesting article!
Reviewer 2 Report
There are still some minor grammatical errors spotted in the text, rest all comments have been addressed.
Reviewer 3 Report
This paper is ok in its current version for me
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
The paper applies the DIRNet deep learning technique on crime data from NY and using the complaint data from the 311 dataset as input. The paper claims to achieve better results than other two methods from 2017, but besides this there is no greater discussion or experiments leading to other conclusions.
Some things that could be improved about the paper are the following:
- The datasets are very poorly described, it is not explained what attributes they have and how they use them, for example explaining if they only use geographic position or how they use time.
- Neither is it explained for how long a prediction is valid, that is, a risk surface per day per hour is predicted?
- Something that I did not understand (because it is not well described) is the time window that appears in figure 7, does that mean that they use that amount of days to train in the past? If so, it would be good for you to comment on whether the method should be retrained every time the time window is changed or is capable of adapting it without retraining.
- Other dimensions could be compared in addition to the F1 score, for example the execution time (where the random forest probably wins)
- Some of the things that are said in 5.5 are quite interesting and it would have been good if they included them in the study, for example knowing which categories of complaints are the ones that most influence the appearance of crimes.
Author Response
Thank you for the helpful and insightful comments. We appreciate all of them very much. Our major revised text is highlighted in yellow in the manuscript. Please check out the attached responses to the comments.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The paper presents a deep-learning driven method for crime prediction using open datasets provided by NYC. The authors have tried to overcome of data imbalance and provide a comparative analysis of various machine learning methods. The authors should however address the following comments:
- Figure 1- Please increase the size of the legends to improve readability.
- Figure 2 – Please, provide more details in the caption as well the legend. (Explicitly mention about the crime types)
- Line 133-136: Please, reword the problem formulation section and provide further details of your choice of under sampling of 0’s and oversampling of 1’s to deal with data imbalance problem?
- Section 5.1: Provide details on what methods have been adopted to optimize the parameters of the deep neural architecture.
- Reword the captions of Figure 7, Figure 8 and summarize the key outcomes
- In the discussion section, it would be nice to have any specific examples of correlation of crimes with geographical neighbourhoods, as well as do you see a peak in crimes at specific time period ?
- There are lot of grammatical issues throughout the manuscript, please revise accordingly.
Author Response
Thank you for the helpful and insightful comments that have made significant changes to the manuscript as a result. We appreciate all of your comments very much. Our major revised text is highlighted in yellow in the manuscript. Please check out the attached responses to the comments.
Author Response File: Author Response.docx
Reviewer 3 Report
I read the article with great interest. The paper deals with the issue of forecasting the occurrence of crimes by means of an innovative deep learning model.
I believe that the article requires only a few improvements, namely:
1) The formatting of the article should be improved, including the references in the text,
2) The section on literature should be expanded and more research should be referred to
3) The conclusions should indicate the research limitations, the direction for future research and the implications of the research carried out.
4) The introduction should also be more focused on the profile of the journal. What is the link between this research and the smart city concept?
Author Response
Thanks for your encouraging feedback. We found the comments from you quite useful and have made significant changes to the manuscript as a result. We appreciate all of your comments very much. Please check out the attached responses to the comments.
Author Response File: Author Response.docx
