Modeling Recidivism through Bayesian Regression Models and Deep Neural Networks
Round 1
Reviewer 1 Report
The paper compares the performance of 3 different methods of statistical analysis; the logistic regression model, Cox regression model, and the cure rate model. The study of risk factors for recidivism of crimes, the relapse to committing a crime or return to criminal activity, is the the focus of this paper. The parameters of these models were estimated from a Bayesian point of view. The presentation of the different statistical methods and the related characteristics, when applied to these kind of data are well presented and discussed. When the aim of the study is the recidivism prediction, the deep neural network’s shows a better performance, when compared to the Cox proportional model and the random forest method. The methods compared are widely used in the analysis of clinical studies and in the epidemiological field. The work therefore takes on a general interest and is useful for researchers in many applied disciplines. I recommend the publication in the present form.
Comments for author File: 
Comments.docx
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this manuscript, the authors compare three common statistical approaches for modeling recidivism:
- Logistic binary regression model.
- Cox regression model.
- and the cure rate model.
The parameters of these models were estimated from a Bayesian point of view. Moreover, the authors discuss the advantages and disadvantages of each model and the availability of different statistical programs to make these analyses. The Cox proportional hazard model (CPH) and random survival forest (RSF) were used to compare the results with the DeepSurv predictions.
Using a real dataset that corresponds to a cohort of individuals, which consisted of men convicted of sexual crimes against women in 1973 in England and Wales. The authors, in this paper, show that the logistic regression model tends to give more precise estimations of the probabilities of recidivism both globally and with the subgroups considered. The prediction results show the deep neural network’s superiority compared to the Cox proportional model and the random survival forest.
I would like to recommend the manuscript for publication in Mathematics if the authors could revise the manuscript in the light of the following minor comments:
- Please revise the references' order carefully, please note that the order of references should be as it first appears in the text.
- In references, there are some punctuations missing (for example, Ref 5). Please check and correct it carefully.
- Line 144: “de” to “of”.
- Two lines before Line 154: “la” to “the”.
- I would suggest that make the axis of Figure 1 a little bigger and use a better resolution Figure.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
After reading the manuscript, I have to say that the manuscript can be accepted in the Journal after major revision. The manuscript needs a serious reconstruction prior publication. Please, find some general and detailed comments below.
- In the abstract, the main aims of the study are missing.
- The organisation of the main chapters is not proper and must be improved. Introductions and conclusion are too long. Some sentences or parts can be moved in other chapters or subchapters. The same comment can be applied for the conclusion. Those chapters must be more concise.
- In general, there is a low number of reference in the whole manuscript. In the subchapter "Logistics regression model" there is no any reference.
- The authors have to avoid repeating the statements across the manuscript.
For me as an environmental scientist is very interesting to see applicability of those methods using completely different datasets that corresponds to a cohort of individuals but in the same time it is not quite clear what is the novelty after using those methods. I would like to ask the authors to explain more clearly what is the benefit of using those methods and what kind of information they have got after using solely the old datasets.
Thus, I recommend this version of the manuscript be returned to the authors allowing them to address the concerns I have expressed in this review. If authors decide to resubmit the improved version of the manuscript, I feel confident that they can address my concerns adequately.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
After reading the revised version of the manuscript, I can say that the authors didn't concern my comments adequately. The authors made only some "cosmetic" changes even I wrote in my previous comments: "the manuscript needs a serious reconstruction prior publication". The new version of manuscript has no conclusion. This chapter must be included in the manuscript.
Thus, I recommend the revised version must be returned once again to the authors and improved accordingly.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 3 Report
The last revised version is improved, significantly. I appreciate the all efforts the authors put in and improved the manuscript accordingly.
