By Gargi Datta https://orcid.org/0000-0002-1314-7824

Review Details

---

Reviewer has chosen not to be Anonymous

Summary of paper in a few sentences:

The paper discusses a way to identify Covid-19 infections from serum samples using machine learning on the results of Raman spectroscopy. They were able to show how 8 different machine learning algorithms performed on classifying Covid-19 using 10-fold CV on 250 raman spectral samples. They also investigated stacking ensemble algorithms and compare the performance of the ensemble to the weak learners.

Reasons to accept:

The paper investigates the efficacy of machine learning to predict a Covid-19 infection from serum data. Raman spectral information can be used to predict Covid-19 and other diseases, and ensemble methods could be an auxiliary tool for clinical diagnosis of Covid-19. Studies like this are very important for advancing machine learning techniques to augment clinical decision making.

Reasons to reject:

I have some questions and concerns about the methodology used in the paper:

Major concerns

1. The sample size used for training the algorithms is very small (250 samples). I worry that with the more complicated methods (for example, a multi-layered perceptron), there is significant overfitting. The authors have used feature selection and extraction techniques to mitigate the high dimensionality issues. They have also used cross-validation to help with overfitting. But, there isn't an independent test set to get performance metrics on. Ideally, metrics on a separate dataset would be reported, but if that is not available, I would suggest the authors set aside say, 10% of the data as an independent test set and evaluate their methods on that dataset in addition to 10-fold CV.

2. In my opinion, the authors report too many metrics in thee results section and the significance of their work is lost a little. I would suggest the authors re-structure the results section and focus on a select few metrics. Additionally, the authors report p-values but don't mention multiple testing error correction. I suggest the authors report Bonferroni corrected p-values.

3. For PCA, it would be beneficial to show a scree plot of Eigen values.

4. Something that would make this paper stronger would be a discussion on the robustness of different models used.

Minor comments:

1. Some citations are missing. For example, citation for subsembles, stacked ensembles, cross-validation when they are first mentioned.

2. I noticed that the authors used selectivity in some of their text. Did they mean specificity?

3. Was SNV done before or after feature selection? If it was done after, what is the reasoning behind that?

4. How many initial features were there before feature selection? Please add that information either in methods or results section.

Nanopublication comments:

Further comments:

By Prableen Kaur https://orcid.org/0000-0003-4912-527X

Review Details

---

Reviewer has chosen not to be Anonymous

Summary of paper in a few sentences:

The manuscript is related to the use of machine learning methods for COVID19 detection.

Reasons to accept:

The idea seems to be interesting. Different machine learning techniques have been used to explore distinct performance metric.

Reasons to reject:

No reason

Nanopublication comments:

Further comments:

The manuscript is related to the use of machine learning methods for COVID19 detection. The idea seems to be interesting. Different machine learning techniques have been used to explore distinct performance metric. However, to maintain the general interest of the reader, the manuscript need to be revised as per the following suggestions:

- First of all, more background work need to be addressed. The novelty and worth of this work need to be reflected in the introduction section.

- There should be a separate related works section.

- As the whole story revolved around the use of machine learning applications. Therefore, a brief paragraph regarding the basics and foundation of the machine learning techniques need to be added in the methodology section. The general applciations of the machine learning techniques need to be briedly highlighted. For this authors may read and refer the following manuscripts.

o Kim, Gi Bae, et al. "Machine learning applications in systems metabolic engineering." Current opinion in biotechnology 64 (2020): 1-9.

o Kaur, Prableen, and Manik Sharma. "Analysis of data mining and soft computing techniques in prospecting diabetes disorder in human beings: a review." Int. J. Pharm. Sci. Res 9 (2018): 2700-2719.

o Kaur, Prableen, and Manik Sharma. "A Smart and Promising Neurological Disorder Diagnostic System: An Amalgamation of Big Data, IoT, and Emerging Computing Techniques." Intelligent Data Analysis: From Data Gathering to Data Comprehension (2020): 241-264.

o Arora, Sankalap, Manik Sharma, and Priyanka Anand. "A novel chaotic interior search algorithm for global optimization and feature selection." Applied Artificial Intelligence 34.4 (2020): 292-328.

o Gan, Lirong, Huamao Wang, and Zhaojun Yang. "Machine learning solutions to challenges in finance: An application to the pricing of financial products." Technological Forecasting and Social Change 153 (2020): 119928.

o Saadatmand, Mohammadsaleh, and Tuğrul U. Daim. "Technology Intelligence Map: Finance Machine Learning." Roadmapping Future: Technologies, Products and Services (2021): 337-3

- The results presented in the figure 2 need to be discussed in mreo detailed manner.

- The strength and limitation of this work need to be clearly addressed.

Reviewer has chosen to be Anonymous

Summary of paper in a few sentences:

Authors proposed classifier comprised of a deep learning predictive meta algorithm for to predict COVID-19 infection. Algorithmic details, datasets and results are good.

Reasons to accept:

Paper is well written. Results are validated. Solution to the problem undertaken is needed today.

Reasons to reject:

Nil

Nanopublication comments:

Further comments: