Comparative Analyses of Selected Neural Networks for Prediction of Sustainable Cementitious Composite Subsurface Tensile Strength

Round 1

Reviewer 1 Report

The article “Comparative analyses of selected machine learning algorithms for prediction of sustainable cementitious composites subsurface tensile strength” (Manuscript ID: applsci-2324303) presents a way of predicting the subsurface tensile strength of mortars with addition of granite fines using Schmidt Hummer test and analysing results  by artificial neural network. Article is interesting but in my opinion some aspects need to be explained wider or corrected as follows:

11.       Line 80: Authors should also mention the EN 13971 standard which describes the way of using NDT methods in estimating the compressive strength of concrete in construction.

22.       Table 1 – needs a proper caption

33.       Is the composition of mortar from Table 1 is being used on construction site?  This is a mortar not concrete.

44.       Line 157: MPa/s

55.       Figure 4 and 5: What are the uncertainties of received values?

66.       Line 220: W/C ratio is divided from C and W, it should not be taken as an additional parameter.

77.       Table 3 is missing.

88.       After performing tests using Schmidt Hammer according to EN 12504-2 you receive one value (one value from nine valid readings) please explain why you took, as I understand, all partial values to the ANN. Same is for compressive strength where you should take an average.

99.       Line 254: Please explain how did you received 192 data sets when you have tested four mortar compositions in two various curing time.

110.   Point 6: When you compared received results of models did you included the uncertainties of test methods? For pull-off method those are really high?

111.   Line 463-466:  I agree, but only when you are using mortars that compositions are in the range of those tested.

112.   Line 468 “properties of concrete”: I agree but you have to create a new model for concrete basing on results of new tests.

113.   Line 472-476: Limitations should include also a fact that mortars, not concrete have been tested.

Author Response

We are very grateful for all valuable comments. We believe that the revised version will meet the expectation of the reviewer and the editor. Please see the attachment for the responses to the Comments. We marked blue the improvements in the Manuscript file.

Author Response File: Author Response.pdf

Reviewer 2 Report

1. Overview of proposed work in Section 1 should be elaborated. Please summarize the technical contributions and novelty of proposed work as well.

2. Section 2.2.3 should not just focus on reviewing the existing works of using ANN for prediction. It should also provide more relevant descriptions of the proposed methodology. For instance, what are the input variables to be fed into the ANN? What are the output variable to be predicted? What are the activation functions used?

3. It is also not clear what are the datasets used by the authors to train ANN? Please provide more detailed description about the dataset used. Any data preprocessing is needed?

4. It is not necessary to present the results in separate sections of 5.1 to 5.8 because they only involve the changes of hyperparameters, i.e., training algorithms and numbers of hidden layers of ANN. In fact, authors should present all results into single section for readers to have better idea about the prediction performance of ANN under different parameter settings. A table should be used to summarize the quantitative results of ANN with different hyperparameter settings too. 

5. Results in Figures 15 and 16 look quite confusing and hard to interpret. Authors should consider more effective way to visualize the results in order to convey the intended message. Descriptions provided in legend are not clear as well. Better write the full name for better descriptions. 

6. Prediction performance of ANN should be compared with other ML algorithm such as linear regression, random forest, SVM and etc to verify the strength of proposed method. 

7. Any further studies have been conducted to investigate which input parameters have more contributions in the prediction of output response? Please clarify. 

Author Response

We are very grateful for all valuable comments. We believe that the revised version will meet the expectation of the reviewer and the editor. Please see the attachment for the responses to the Comments. We marked blue the improvements in the Manuscript file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Accept in present form

Reviewer 2 Report

The quality of revised manuscript has been improved. Authors have addressed most of the comments given in earlier review process. This manuscript can be accepted for publication.