Estimation of Obesity Levels with a Trained Neural Network Approach optimized by the Bayesian Technique

Round 1

Reviewer 1 Report

Review Report

 

 

The present study aimed to predict the obesity levels based on physical activity and eating habits using a trained neural network model. They also employed various classify algorithms to identify the strongest features associated with obesity. Overall, study seems to utilize a complex machine learning method and identified the already known predictors of obesity, that are nothing new to the field. However, the study methods can be indeed useful to discover predictors for other complex metabolic diseases where we lack strong disease classifiers. 

 

I have various major concerns that needs to be addressed in revised version of the manuscript as stated below –

 

1.     What is the need to use a complex machine learning method “neural network” to predict obesity levels when simpler methods already exist – such as SVM, decision tress, random forest, gradient boost methods etc.? It can be very data and time intensive and need advanced resources than simpler methods. Authors needs to address this in their manuscript.

 

2.     Obesity Type 1 group has very low samples for training and testing sets that may not well correlate with model accuracy. What makes them to be included in the predictions of obesity levels with very small sample number?

 

3.     What do authors mean by insufficient weight? Do they mean underweight? If so, it should be corrected in the manuscript.

 

4.     What are the criteria to classify different obesity levels from insufficient weight to obesity type III? Is it body mass index (BMI)? What are the BMI ranges in different groups?

 

5.     How did authors choose the number of hidden layers to be one. Did they try to increase the number of hidden layers gradually? Does that improve or have no effect on the performance of the model?

 

6.     Somewhere in results authors state – “According to the SD values, it is concluded that the model is robust”. However, SD between the trials is very high for obesity type I and type II (Table 4)? Any comments on that are highly appreciated.

 

7.     Figure 2 resolution is very poor, and nothing can be visualized clearly. Such bad quality figures are heavily discouraged.

 

8.     How the frequency of vegetable consumption correlates with obesity risk needs some explanation in the manuscript. Are their specific vegetables that needs to be avoided/promoted etc. and that can be recommended?

 

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Yagin and colleagues developed a trained neural network to predict obesity with a high accuracy in this manuscript. Moreover, the authors also identified eating habits and physical activity were the most critical factors for obesity estimation. Overall, this is very interesting topic, but a few questions need to be addressed.

1. It is not surprising that both eating habits and physical activity are able to affect the prevalence of obesity. For instance, people who eat more and have less physical activity are more likely to have obesity. That’s to say the conclusion of this manuscript is common sense in daily life. However, from another angle, the conclusion supports the complex analysis developed by the authors. Authors should discuss it in the manuscript. By the way, the title can be optimized without telling based on eating habits and physical activity states. In fact, the authors tested 17 characteristics. One more thing, it is hard for readers to get the conclusion that eating habits and physical activity are highly involved. The authors need to highlight somewhere in the results part.

2. The trained neural network appears working well with a high accuracy in the dataset used in the manuscript. Did the authors test or validate it in other datasets?

3.  The Introduction part talks too much about epidemics and other aspects of obesity itself, which is not necessary for this manuscript. The authors need spend more words on the significance of appropriate analysis to estimate obesity, or introduction upon various statistical analysis methods that were used in this manuscript.

4. Figure are very blurry. I can’t read Figure 2.

5. The language of this manuscript needs to be further checked/improved by a native speaker in this field.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Authors have utilized neural networks to estimate obesity levels based on eating habits and physical activity states. This work is potentially important and would be of interest to the readers. 

The manuscript is well written, I have no further comments. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

The manuscript by Fatma Hilal Yagin et al. entitled "A Machine Learning-Based Approach for Estimating Obesity Levels: A Study from Colombia, Mexico, and Peru" presents a study on using machine learning techniques to predict obesity levels in individuals from Colombia, Mexico, and Peru. The authors use a dataset that includes 17 characteristics and 2111 records, and a neural network model with one hidden layer is developed using the Keras library. The model's performance is evaluated using F1-score, accuracy, and other measures.

Overall, the paper provides a thorough and well-structured description of the research conducted and the results obtained. The use of machine learning techniques to predict obesity levels is an important and relevant topic, and the study provides valuable insights and contributions to the field. However, there are also some areas that could be improved upon in the paper.

 

Major comments:

1.      The authors should provide more information on the dataset used in the study, such as the distribution of participants across different age groups, and the distribution of participants across different countries. Information on the demographics of the participants would be helpful in understanding the generalizability of the results.

2.      The materials and methods section is well-written and provides clear information on the dataset and experimental analysis. However, it would be helpful to provide more details on the survey used to collect the data, including the specific questions asked and the response rate.

3.      The authors should provide more information on the feature selection algorithms used in the study, such as the specific parameters used and how they were chosen.

4.      The authors should provide more information on the neural network model used in the study, such as the specific architecture of the model and how the hyperparameters were chosen. For example, it would be helpful to have a separate section on the feature selection and hyperparameter optimization methods used.

 

Minor comments:

1.      The comparison of the results of this study to previous studies in the literature is limited and could be expanded upon. A more thorough comparison of the results to other studies in the field would provide a better understanding of how this study fits into the existing literature.

2.      The introduction could provide more background on the specific countries (Colombia, Mexico, and Peru) included in the dataset and how they relate to the overall problem of obesity.

 

3.      The tables and figures should be labeled more clearly and include a caption that provides a brief description of the data presented.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Authors quickly performed the revision based on all suggestions except that the resolution of figure 2 is still not a very good quality and can be improved to a better quality using graphPad prism or R.

Author Response

Reviewer #1:

Q1. Authors quickly performed the revision based on all suggestions except that the resolution of figure 2 is still not a very good quality and can be improved to a better quality using graphPad prism or R.

Answer 1: Thank you very much. Figure 2 has been redrawn and revised. In addition, the revised figure was added to the text.