Compressive Strength Estimation of Waste Marble Powder Incorporated Concrete Using Regression Modelling

Round 1

Reviewer 1 Report

This article is very interesting because it examines the use of machine learning (ML) capabilities to predict the compressive strength of concrete incorporating WMP for future use.

Abstract, is self-explanatory, explaining the background, aims and benefits of the research, the methods used and the results of the research.

Introduction, especially in the background, the Author has very clearly outlined the problems, objectives and benefits of this paper however, the Author needs to reaffirm the significance of the research or the urgency of this research.

The method has also been described quite clearly in data collection, but it needs to be emphasized again in analyzing the data in answering the problem formulation.

The results and discussion are strong enough to describe that the experimental results show that random forest is the best model for predicting compressive strength with an R2 value of 0.926 and an MAE of 1.608. Furthermore, SHAP analysis and VIF analysis showed that the best regression model ability was achieved in optimizing the use of marble powder as a partial substitute for cement in concrete.

In terms of grammar, spelling is very good, it just needs to be fixed, the same vocabulary appears, maybe you need to find another vocabulary or paraphrase it.

Some of the images are still small and difficult to read and need to be clarified, as well as tables and figures made clearer, complete with their sources.

References are complete with updated sources, both books, articles and journals. You only need to check again whether it is in accordance with the template and connect with the citations in the article.

Author Response

The Authors' Responses to the Reviewer's Concerns on Manuscript ID: coatings-2088993

Title: COMPRESSIVE STRENGTH ESTIMATION OF WASTE MARBLE POWDER INCORPORATED CONCRETE USING REGRESSION MODELLING

Dear Reviewer 1,

We thank you for sparing your precious time to read and review the manuscript. All your valuable comments and reviews have been taken into consideration. We have thoroughly revised the manuscript to address the reviewer's concerns and incorporate their suggestions. Improvements and changes in the manuscript are done accordingly. The changes in the manuscript have been shown using the inline formatting scheme. 

We hope to hear positively from the editor and reviewers.

Sincerely,

Manpreet Singh

Priyankar Choudhary

Anterpreet Kaur Bedi

Saurav Yadav

Rishi Singh Chhabra

Reviewer comments: 

1. This article is very interesting because it examines the use of machine learning (ML) capabilities to predict the compressive strength of concrete incorporating WMP for future use.

Abstract, is self-explanatory, explaining the background, aims and benefits of the research, the methods used and the results of the research.

Action taken: We thank the reviewer for the comment. We have again revised the entire manuscript so that minor errors are also removed. Language and grammar have been rechecked and corresponding changes have been made.

2. Introduction, especially in the background, the Author has very clearly outlined the problems, objectives and benefits of this paper however, the Author needs to reaffirm the significance of the research or the urgency of this research.

Action taken: We thank the reviewer for his valuable comment. The last paragraph of the introduction section emphasizes on the need for the present research in the current scenario. In addition, the authors have added the following in the introduction section in order to reaffirm the significance of the research as is shown below:

• “...With the advancement of various soft computing techniques, data handling capabilities of researchers have increased and are more efficient now over conventional ways. As a result, many algorithms have gained popularity in due course of time. However, a detailed comparison between these algorithms still remains less explored and need further study. Use of machine learning (ML) algorithms can provide a reliable mix design for industry. In the long run, the Indian Standard codes can also be updated with marble dust parameter…”
• “...Although a huge number of experimental studies have been carried out for investigating possible effects of WMP on concrete, however, there is still a lack of in-depth understanding on use of WMP in concrete...”

3. The method has also been described quite clearly in data collection, but it needs to be emphasized again in analyzing the data in answering the problem formulation.

Action taken: The comment has been noted. For the current work, data has been collected by one of the authors in his previous research that was purely experimental (Paper: M. Singh, K. Choudhary, A. Srivastava, K.S. Sangwan, D. Bhunia, ‘A study on environmental and economic impacts of using waste marble powder in concrete,’ J. Build. Eng., 13 (2017), pp. 87-95). The same data has been used in the current research. The authors have now tried to explain the same in the Data collection section in an improved manner as shown:

“...Three water binder ratios 0.35, 0.4 and 0.45 were considered with two mix proportion designs for each of the ratios. Two variations in dosage of admixture and corresponding slump values were considered for each of the mix designs. Once the dosage of superplasticizer was kept constant for increasing replacement percentage of marble slurry which resulted in increasing slump values and for the other case dosage of admixture was increased for increasing replacement percentage of marble slurry for keeping the slump values constant. These variations provided 12 different mix designs. Finally, for each mix design 5 different marble slurry percentage replacement levels were considered, resulting in a total of 60 mix designs. 12 concrete cube samples for each variation were casted and tested thus generating a data set of 720 values…”

In addition, the same has been emphasized again in the results section (Section 4.1) as mentioned below:

“....As described earlier, the data for the current work was collected experimentally by Singh et. al for 12 different sets of design mixes, with each mix considering 5 different marble slurry percentage replacement levels, resulting in 60 variations. A set of 720 data points was generated by casting 12 concrete cubes for each of the 60 variations. From the entire data of 720 samples, 70% is considered for training, while the rest 30% is used for testing purposes.….”

4. The results and discussion are strong enough to describe that the experimental results show that random forest is the best model for predicting compressive strength with an R2 value of 0.926 and an MAE of 1.608. Furthermore, SHAP analysis and VIF analysis showed that the best regression model ability was achieved in optimizing the use of marble powder as a partial substitute for cement in concrete.

Action taken: We thank the reviewer for the comment. Further, in support of the application of RF technique, we have added a few points in the result section as shown below:

“...From the above table, it can be observed that different performance measures consider different ML models as the best performing. The R2 score, MAE, MSE and MAPE values are the best for RF model. On the other hand, RMSE and MBE values are best shown for MLR, whereas T_stat is best for GB modelling technique. Since the RF model is best for the majority of the performance parameters, and the other three parameters, i.e RMSE, MBE and T_stat do not show any significant best performing model, hence RF was considered for further analysis. Thus, it can be seen that MLR is the least preferred model for regression since all the performance measures except RMSE and MBE are least preferred, thus leading to the need of more complex models for prediction. Also, with exponentially increasing data in the current scenario, the applicability of MLR becomes minimal ....”

5. In terms of grammar, spelling is very good, it just needs to be fixed, the same vocabulary appears, maybe you need to find another vocabulary or paraphrase it.

Action taken: We acknowledge the comment from the reviewer. We have thoroughly revamped the paper to improve its readability. Editing has been carried out in each section to improve the readability and understanding of the text  

6. Some of the images are still small and difficult to read and need to be clarified, as well as tables and figures made clearer, complete with their sources.

Action taken: The authors have noted the comment. We have now added the images in higher quality so that they are clearer to study and understand. All the subimages in figure 3 have been rescaled such that the data values are more clearly visible. Figure 4 has now been entirely changed and separate graphs showing fluctuations in error have been added for each model. The results are more clearly visible. Table 2 has now been rotated so that it is more comfortable for the readers to read. Also, all the figures have been drawn by the authors, hence, no sources need to be mentioned in the paper.

7. References are complete with updated sources, both books, articles and journals. You only need to check again whether it is in accordance with the template and connect with the citations in the article.

Action taken: The authors have cross checked the references to ensure their accordance with the template, and connections with the citations in the article.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper the authors investigate the effect of waste marble powder on the compressive strength of concrete using machine learning. Overall, the article is quite well structured; the objectives and the methodology are well defined and the conclusions are concise. Please, find below some minor comments that I hope you can take in consideration:

- In the abstract you should avoid the use of acronyms. For example, MAE and VIF are not even written in their full length;

- In introduction, some concept are redundant (i.e. line 40-41). Please double check.

- Line 42 to 61: a little too generic and out of the scope of the paper. Remove it, or try to summarize, please;

-In table 2, you mentioned “admixture”. Is it the same than plasticizer? If yes, please try to use the same term to avoid misunderstanding;

-Table 2 is also very small and difficult to read. Please, increase the font size;

-I suggest you to relate the results of table 5 with the description of table 4 (i.e. what does mean that RF has the lowest value of MAE? Why MLR the lowest RMSE? And so on…

- since you used a lot of acronysms, I suggest you to list them at the begin or at the end of the paper.

Author Response

The Authors' Responses to the Reviewer's Concerns on Manuscript ID: coatings-2088993

Title: COMPRESSIVE STRENGTH ESTIMATION OF WASTE MARBLE POWDER INCORPORATED CONCRETE USING REGRESSION MODELLING

Dear Reviewer 2,

We thank you for sparing your precious time to read and review the manuscript. All your valuable comments and reviews have been taken into consideration. We have thoroughly revised the manuscript to address the reviewer's concerns and incorporate their suggestions. Improvements and changes in the manuscript are done accordingly. The changes in the manuscript have been shown using the inline formatting scheme. 

We hope to hear positively from the editor and reviewers.

Sincerely,

Manpreet Singh

Priyankar Choudhary

Anterpreet Kaur Bedi

Saurav Yadav

Rishi Singh Chhabra

Reviewer comments: 

In this paper the authors investigate the effect of waste marble powder on the compressive strength of concrete using machine learning. Overall, the article is quite well structured; the objectives and the methodology are well defined and the conclusions are concise. Please, find below some minor comments that I hope you can take in consideration:

1. - In the abstract you should avoid the use of acronyms. For example, MAE and VIF are not even written in their full length;

Action taken: The comment has been noted. We have now revised the abstract to avoid the use of any acronyms. All the acronyms used in the abstract, namely WMP, ML, MAE, SHAP and VIF, have been removed and full forms have been added instead. In addition, a list of acronyms has been added at the end of the paper to make it convenient for the readers.

2. - In introduction, some concepts are redundant (i.e. line 40-41). Please double check.

Action taken: The authors have noted the comment. The authors have revised the introduction section entirely and observed that the statement “Moreover, as far as cement is concerned, it contributes to about 8% of the world’s carbon dioxide emissions.” was already added at the end of the introduction. Hence, it has now been removed.

3. - Line 42 to 61: a little too generic and out of the scope of the paper. Remove it, or try to summarise, please;

Action taken: We understand the reviewer’s concern. The authors understand that lines 42-61 are quite generic and hence, should not be included in the manuscript. The paragraph has been revised and removed from the manuscript. A statement for summarising the generalised concept has been added as shown below:

“..With the advancement of various soft computing techniques, data handling capabilities of researchers have increased and are more efficient now over conventional ways. As a result, many algorithms have gained popularity in due course of time. However, a detailed comparison between these algorithms still remains less explored and need further study. Use of machine learning (ML) algorithms can provide a reliable mix design for industry. In the long run, the Indian Standard codes can also be updated with marble dust parameters..”

4. -In table 2, you mentioned “admixture”. Is it the same than plasticizer? If yes, please try to use the same term to avoid misunderstanding;

Action taken: Admixture and plasticizer refer to the same material. The authors have now used the term superplasticizer in the entire paper to maintain consistency of the text. The same has been changed in Figures 6 (VIF analysis for RF) and 7 (SHAP analysis) as well.

5. -Table 2 is also very small and difficult to read. Please, increase the font size;

Action taken: The comment has been taken care of. The table has now been rotated so that it is more comfortable for the readers to read.

6. -I suggest you to relate the results of table 5 with the description of table 4 (i.e. what does mean that RF has the lowest value of MAE? Why MLR the lowest RMSE? And so on…

Action taken: The comment has been noted. In response, the authors have added a detailed explanation relating results of Table 5 with the description of table 4 as shown below:

“...From the above table, it can be observed that different performance measures consider different ML models as the best performing. The R2 score, MAE, MSE and MAPE values are the best for RF model. On the other hand, RMSE and MBE values are best shown for MLR, whereas Tstat is best for GB modelling technique. Higher R2 score for RF indicates that the model best fits the dataset compared to the rest. Further, the model shows least fluctuation in errors, as is indicated by MSE, MAE and MAPE values. This shows that there is minimal variance in residuals for RF model in comparison to the other ML models. Since the RF model is best for the majority of the performance parameters, and the other three parameters, i.e., RMSE, MBE and Tstat do not show any significant best performing model, hence RF was considered for further analysis. Thus, it can be seen that MLR is the least preferred model for regression since all the performance measures except RMSE and MBE are least preferred, thus leading to the need of more complex models for prediction. Also, with exponentially increasing data in the current scenario, the applicability of MLR becomes minimal….”

7. - since you used a lot of acronyms, I suggest you to list them at the beginning or at the end of the paper.

Action taken: The comment has been noted and a list of abbreviations used in the manuscript has been added at the end of the manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

This manuscript investigates the compressive strength of concrete mixture with waste marble power by incorporating machine learning. The research topic may contribute to the regional or global environment and sustainability, and the research method is inceptive in Civil Engineering. I have a few major comments as follows:

Technical:
1. Lines 3-4, Abstract. This should be included in the “Background” or “Introduction” section, not in the “Abstract.”

2. Lines 32-33, Introduction. There appear to be some potential LEED (i.e., Leadership in Energy and Environmental) values in this research topic by reusing construction waste and also reducing the environmental impact.

3. Lines 153-165, Methodology. It appears that all 720 instances were designed in the previous experimental tests by Singh et al. Please confirm or provide more clarifications.    

4. Lines 188-189, Modeling. What is the corresponding process time for each modeling technique?

5. Figure 3. Advise the authors to change the figures with the same range and scale in both the horizontal and vertical axes, which will visually show the discreteness between the predicted and actual values. For instance, the current vertical scale appears to be much narrower than the horizontal scale.

6. Figure 4. This figure should be modified to clearly differentiate the results from all modeling methods.

7. Table 5. Are all the performance measures supporting the same conclusion? If not, which ones provide the best prediction? Can any conclusions be drawn from this?

General:

8. Table 4. The T_stat in Table 4 is different from the one in Table 5. Please revise.  

9. All acronyms shall be explained in the first appearance and be utilized afterward. For instance, the first appearance of KNN is in Line 178, not Line 212. Please update all.

10. Line 384. “SHapely” should be “Shapely.”

Author Response

The Authors' Responses to the Reviewer's Concerns on Manuscript ID: coatings-2088993

Title: COMPRESSIVE STRENGTH ESTIMATION OF WASTE MARBLE POWDER INCORPORATED CONCRETE USING REGRESSION MODELLING

Dear Reviewer 3,

We thank you for sparing your precious time to read and review the manuscript. All your valuable comments and reviews have been taken into consideration. We have thoroughly revised the manuscript to address the reviewer's concerns and incorporate their suggestions. Improvements and changes in the manuscript are done accordingly. The changes in the manuscript have been shown using the inline formatting scheme. 

We hope to hear positively from the editor and reviewers.

Sincerely,

Manpreet Singh

Priyankar Choudhary

Anterpreet Kaur Bedi

Saurav Yadav

Rishi Singh Chhabra

Reviewer comments: 

This manuscript investigates the compressive strength of concrete mixture with waste marble powder by incorporating machine learning. The research topic may contribute to the regional or global environment and sustainability, and the research method is inceptive in Civil Engineering. I have a few major comments as follows:

Technical:

1. Lines 3-4, Abstract. This should be included in the “Background” or “Introduction” section, not in the “Abstract.”

Action taken: The comment has been noted and acted upon. The authors observed that the same statement has already been added in the Introduction section as shown below:

“...Numerous researches have been performed in the past in order to introduce newer materials as replacement of cement. However, the conventional approach of relying solely on laboratory test data is quite costly and inefficient. One requires an impractically huge number of controlled testing to reach a reasonable conclusion and thus roll out innovations in the construction industry ....”

Hence, the authors have removed lines 3-4 (Abstract) to remove redundancy.

2. Lines 32-33, Introduction. There appear to be some potential LEED (i.e., Leadership in Energy and Environmental) values in this research topic by reusing construction waste and also reducing the environmental impact.

Action taken: The authors appreciate the reviewer's comment. This research is an extension of a project funded by the Department of Science and Technology on utilisation of the waste marble powder as a potential replacement of cement in concrete. Singh et al.  have conducted a detailed study on environmental impact analysis of use of this material as a partial replacement of cement and sand. In the study, authors have taken into account the effect on agricultural land and water depletion as impacted areas. A brief discussion of the same was added in the revised manuscript. 

Paper: M. Singh, K. Choudhary, A. Srivastava, K.S. Sangwan, D. Bhunia, ‘A study on environmental and economic impacts of using waste marble powder in concrete,’ J. Build. Eng., 13 (2017), pp. 87-95

3. Lines 153-165, Methodology. It appears that all 720 instances were designed in the previous experimental tests by Singh et al. Please confirm or provide more clarifications.

Action taken: The comment has been noted. For the current work, data has been collected by one of the authors in his previous research that was purely experimental (Paper: M. Singh, K. Choudhary, A. Srivastava, K.S. Sangwan, D. Bhunia, ‘A study on environmental and economic impacts of using waste marble powder in concrete,’ J. Build. Eng., 13 (2017), pp. 87-95). The same data has been used in the current research. The authors have now tried to explain the same in the Data collection section in an improved manner as shown:

“...Three water binder ratios 0.35, 0.4 and 0.45 were considered with two mix proportion designs for each of the ratios. Two variations in dosage of admixture and corresponding slump values were considered for each of the mix designs. Once the dosage of superplasticizer was kept constant for increasing replacement percentage of marble slurry which resulted in increasing slump values and for the other case dosage of admixture was increased for increasing replacement percentage of marble slurry for keeping the slump values constant. These variations provided 12 different mix designs. Finally, for each mix design 5 different marble slurry percentage replacement levels were considered, resulting in a total of 60 mix designs. 12 concrete cube samples for each variation were casted and tested thus generating a data set of 720 values…”

In addition, the same has been emphasised again in the results section (Section 4.1) as mentioned below:

“....As described earlier, the data for the current work was collected experimentally by Singh et. al for 12 different sets of design mixes, with each mix considering 5 different marble slurry percentage replacement levels, resulting in 60 variations. A set of 720 data points was generated by casting 12 concrete cubes for each of the 60 variations. From the entire data of 720 samples, 70% is considered for training, while the rest 30% is used for testing purposes.….”

4. Lines 188-189, Modeling. What is the corresponding process time for each modelling technique?

Action taken: The processing time for each modelling technique is mentioned below.

From the table, it can be observed that the processing time (in msec) is least for MLR followed by KNN and RF. However, the best model (RF) has been chosen taking into consideration other performance parameters describing better applicability of the model. Also, the processing time taken for each technique may vary based on the specifications of the processor. Since the basis of selecting the best model were the performance in terms of least fluctuation in errors, hence processing time was not added in the manuscript.

5. Figure 3. Advise the authors to change the figures with the same range and scale in both the horizontal and vertical axes, which will visually show the discreteness between the predicted and actual values. For instance, the current vertical scale appears to be much narrower than the horizontal scale.

Action taken: The comment has been taken care of. All the subimages in figure 3 have been rescaled such that the data values are more clearly visible. The vertical scale has been made comparable to the horizontal scale. Also, the font size of values and labels has also been increased for better reading.

6. Figure 4. This figure should be modified to clearly differentiate the results from all modeling methods.

Action taken: The authors strongly agree with the reviewer’s comment. In order to address the comment, figure 4 has now been entirely changed and separate graphs showing fluctuations in error have been added for each model. The results are more clearly visible. In addition, a residual variance value for each model has also been calculated and mentioned in each corresponding graph. Since minimum variation in errors is desirable, it can be observed that RF results in least fluctuations, and hence, is more suited for prediction of compressive strength compared to the rest of the models.

7. Table 5. Are all the performance measures supporting the same conclusion? If not, which ones provide the best prediction? Can any conclusions be drawn from this?

Action taken: The table 5 in the manuscript gives the performance measures for each model. From the table, it can be observed that different performance measures consider different ML models as the best performing. The R2 score, MAE, MSE and MAPE values are the best for RF model. On the other hand, RMSE and MBE values are best shown for MLR, whereas Tstat is best for GB modelling technique. Higher R2 score for RF indicates that the model best fits the dataset compared to the rest. Further, the model shows least fluctuation in errors, as is indicated by MSE, MAE and MAPE values. This shows that there is minimal variance in residuals for RF model in comparison to the other ML models.Since the RF model is best for the majority of the performance parameters, and the other three parameters, i.e., RMSE, MBE and Tstat do not show any significant best performing model, hence RF was considered for further analysis.  The same has been added in the manuscript in support of the conclusions, along with the following:

“...it can be seen that MLR is the least preferred model for regression since all the performance measures except RMSE and MBE are least preferred}, thus leading to the need of more complex models for prediction. Also, with exponentially increasing data in the current scenario, the applicability of MLR becomes minimal….”

General:

8. Table 4. The T_stat in Table 4 is different from the one in Table 5. Please revise.  

Action taken: The comment has been noted and the same has been revised in Table 5.

9. All acronyms shall be explained in the first appearance and be utilised afterward. For instance, the first appearance of KNN is in Line 178, not Line 212. Please update all.

Action taken: The comment has been noted. We have now revised the manuscript such that the acronyms have be explained in the first appearance and utilised afterwards. In addition, a list of acronyms has been added at the end of the paper to make it convenient for the readers.

10. Line 384. “SHapely” should be “Shapely.”

Action taken: The term SHapley Additive exPlanations is written in this manner to explain the corresponding acronym SHAP. (Lundberg, Scott M., and Su-In Lee. "A unified approach to interpreting model predictions." Advances in neural information processing systems 30 (2017).)

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

This manuscript could be accepted.