CBM Gas Content Prediction Model Based on the Ensemble Tree Algorithm with Bayesian Hyper-Parameter Optimization Method: A Case Study of Zhengzhuang Block, Southern Qinshui Basin, North China

Round 1

Reviewer 1 Report

The author well explained the CBM gas content using Tree Algorithm with  Bayesian Hyper-Parameter optimization method.

However, I am hereby sending some comments to improve further, 

1. It is preferable to include the dataset in a separate file so that we can better understand the model's accuracy.

2. Include dataset programming.

3. Compare with another algaritham

 

The work is excellent, but without the dataset, it is difficult for the reviewer to evaluate it. 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The introduction section must be improved. Carbon neutrality is a big theme; the authors do not detail and use only 4 references. It is not clear the importance of the results of the manuscript. 

In the methodology section, there is not enough information to use the tree algorithm. Reference 24 is a part of the selection criteria and authors must include a brief discussion of the quantitative selection of this algorithm. 

Eq. 3 needs a higher explanation about the expected value for f0(x). Close to 0.0001 or close to 0.01? 

Section 3 is a piece of excellent information for the readers. Congrats for it.

Figure 4 is a result. Must be placed in Section 4.

Figures 7 (a) and 8 (a)  have an overestimation of gas contents at lower values, it needs to separate the results into two groups: 0 to 15 m3/t and 15 to 30 m37t, and explain the causes of the new R2 values. One R2 value for 0 to 15 and a second R2 value for 15 to 30. This separation would modify lines 313 to 316 for the 15 % error line in general results.

Figure 12 is very good information. Congrats again.

The Conclusion section must include the main values of the prediction besides the accuracy of BO-GBDT. 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The paper is nicely written and structured and the findings can be of interest to readers of the journal. I can recommend it for publication after I find the below comments addressed. The authors are also invited to perform a double-check of English as I found some minor typos and grammatical issues.
-       Abstract may need some results regarding the models and comparisons.
-       There are two concerns regarding the introduction:
It is well structured but although the literature review can comply with the required level of coverage, I believe using more recent works will increase the background value of the paper. improving the conceptual aspect below works are recommended:
Forecasting of water thermal conductivity enhancement by adding nano-sized alumina particles, Estimation of kinematic viscosity of biodiesel-diesel blends: Comparison among accuracy of intelligent and empirical paradigms, Developing a Hybrid Neuro-Fuzzy Method to Predict Carbon Dioxide (CO2) Permeability in Mixed Matrix Membranes Containing SAPO-34 Zeolite, Prediction of viscosity of biodiesel blends using various artificial model and comparison with empirical correlations, Numerical study on the application of biodiesel and bioethanol in a multiple injection diesel engine.

Also, the novelty and selection of these algorithms should be supported by discussing the weaknesses of previous works, advantages of algorithms, etc.
I would like to know why metaheuristic algorithms like GBDTare compared to conventional models like RF .
-       A more profound discussion is required by linking the findings to the objective and state-of-the-art knowledge.
-       Please make sure that all test conditions and parameters like the number of search agents, iteration, steps, etc. are clearly given so that the readers be able to redo/simulate the work.

- Some important information (topology of networks, activation function, etc.) is lacking. Moreover, the hyperparameters of a network are the topology and training parameters, not the weigh. The hyperparameters are called hyperparameters because they cannot be determined by directly numeric optimization. We do not see any discussion on the method to determine hyperparameters.

- The differences of the errors are not significant for most training algorithms. Since a specific algorithm may obtain different results in different runs, how could the authors prove that the differences are caused by the algorithms themselves instead of the random fluctuation. Are the results obtained by a single run? Or they are the mean/best values after multiple runs?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The queries were answered carefully, it is now ready for publication.