Anomaly Detection Utilizing One-Class Classification—A Machine Learning Approach for the Analysis of Plant Fast Fluorescence Kinetics
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsComments “Anomaly Detection Utilizing One-Class Classification – A Machine Learning Approach for the Analysis of Plant Fast Fluorescence Kinetics.”
Tran, 2024, Stresses 3264479.
The paper describes a machine-learning approach to categorize OJIP-fast fluorescence kinetics measurements into ‘normal’, representing healthy plants and ‘anomalies’. The approach was validated using previously published datasets. The results showed a subgroup of identified anomalies linked to stress-induced reductions in photosynthesis and a meaningful correlation with progression and severity of stress.
Although the paper is well written, it was in a first reading not very clear what the author meant by Type I (false positive) and Type II (false negative) errors and the relation with ‘normal’ and ‘anomaly’.
Paragraph 2.2:
The identification of anomalies was based on Fv/Fm values but what is the threshold to identify a Fv/Fm value as an anomaly? Looking at fig 1, the range in Fv/Fm values is quite high, which is normal in this type of analysis especially from measurements in the field. In this figure a threshold is indicated and data points below this one are considered as outliers. Are these outliers then anomalies or not? What are the underlying mechanisms responsible for these outliers/anomalies?
Paragraph 2.3:
The meaning of the parameter nu is not clear and why varying between 0.01 and 0.5. What is the rationale behind this? Was this analyses performed on the Fv/Fm values? From figure 5 it became clear that the analysis was done on most of the parameters that characterize the OJIP-fluorescence kinetics. It should be handsome to inform the reader of this approach in an earlier paragraph.
Paragraph 2.9:
In the field experiment, plants were subjected to environmental stresses and some of them also to RF-EMF. From the results and the presented analysis, It is not clear to me how a discrimination between the effect of environmental stress on the one hand and the supplementary effect of the RF-EMF on the other hand in the group of anomalies can be made. Please clarify this issue.
Comments for author File: 
Comments.pdf
Author Response
We are grateful of the constructive comments from all reviewers and would like to address them point by point.
Comment: The paper describes a machine-learning approach to categorize OJIP-fast fluorescence kinetics measurements into ‘normal’, representing healthy plants and ‘anomalies’. The approach was validated using previously published datasets. The results showed a subgroup of identified anomalies linked to stress-induced reductions in photosynthesis and a meaningful correlation with progression and severity of stress.
Although the paper is well written, it was in a first reading not very clear what the author meant by Type I (false positive) and Type II (false negative) errors and the relation with ‘normal’ and ‘anomaly’.
Answer: Type I errors are cases where "normal" are misidentified as "anomalies". Type II errors are cases where "anomalies" are misidentified as "normal". These descriptions are mentioned several times both in the text (paragraphs 2.2 and 2.3) and in the legend of Figure 4.
Comment: Paragraph 2.2: The identification of anomalies was based on FV/FM values but what is the threshold to identify a FV/FM value as an anomaly? Looking at fig 1, the range in FV/FM values is quite high, which is normal in this type of analysis especially from measurements in the field. In this figure a threshold is indicated and data points below this one are considered as outliers. Are these outliers then anomalies or not? What are the underlying mechanisms responsible for these outliers/anomalies?
Answer: FV/FM represents the maximum quantum yield of photosystem II. Among the many parameters of chlorophyll fluorescence analysis, FV/FM is one of the most consistent values. It shows minimal variation regardless of differences in cultivar or growth conditions, which otherwise have a large influence on other parameters. It is well known that stress conditions can (but do not necessarily) reduce the values of FV/FM.
For this reason, we have chosen FV/FM as the criterion for screening 'anomalies'. The reasoning is that if a measurement shows a statistically significantly lower value of FV/FM than normal (i.e. an outlier), it is very likely that this is a case of an unhealthy spot, i.e. an "anomaly" (or more precisely: a stress-related “anomaly”). On the other hand, it should be emphasised that FV/FM is a very conservative criterion for anomaly screening due to its high consistency. Many "anomalies" have the same FV/FM values as normal and therefore would be missed by the FV/FM screening. As the matter of fact, we were only able to detect 24 cases of "anomalies" with the FV/FM screening, but 664 cases with our machine learning approach.
We have made changes to the manuscript to highlight these points.
Comment: Paragraph 2.3: The meaning of the parameter nu is not clear and why varying between 0.01 and 0.5. What is the rationale behind this?
Answer: nu represents a lower bound on the fraction of support vectors. Technically, a nu value of 0.05 means that at least 5% of the training samples are support vectors (data points that lie adjacent to the boundary hyperplane). It also represents the upper bound on the fraction of training errors. For example, a nu value of 0.05 means that the model can tolerate training errors of up to 5% (in reality, 11/204 = 5.3% of cases in the training set were identified as "anomalies").
The nu parameter lies between 0 and 1 and should be fine-tuned through cross-validation. In practice, values between 0.1 and 0.5 are common starting points. In our case, nu values greater than 0.5 were not tested because this would only increase the Type II errors, which were already unacceptably large at nu = 0.5. On the other hand, too small nu values are also impractical, as there are only 204 data points in the training set. With the lowest value tested, nu = 0.01, the model will not allow more than 204 x 0.01 = 2 cases of "anomalies".
We have made changes to the manuscript to highlight these points.
Comment: Was this analyses performed on the FV/FM values?
Answer: Yes, the fine-tuning of nu was also done with "anomaly" cases identified from the previous FV/FM screening. Over all values of nu tested, our model successfully identified all 24 cases correctly as "anomalies".
Comment: From figure 5 it became clear that the analysis was done on most of the parameters that characterize the OJIP-fluorescence kinetics. It should be handsome to inform the reader of this approach in an earlier paragraph.
Answer: We agree
Comment: Paragraph 2.9: In the field experiment, plants were subjected to environmental stresses and some of them also to RF-EMF. From the results and the presented analysis, It is not clear to me how a discrimination between the effect of environmental stress on the one hand and the supplementary effect of the RF-EMF on the other hand in the group of anomalies can be made. Please clarify this issue.
Answer: It is beyond the scope of this paper, but in the cited study we examined the effects of RF-EMF on plants under both control and field conditions and found that the adverse effects of RF-EMF were manifested only in the field. Thus, we concluded (with further evidence) that RF-EMF may interfere with plant stress responses. It should be noted that the results are still quite preliminary and further research is needed to clarify this aspect.
As the data used in this paper (which is only part of the data used in our last study) are not sufficient per se for this conclusion, we have decided to remove the mention of this effect as it could cause confusion.
Reviewer 2 Report
Comments and Suggestions for AuthorsDear Editors and Authors, I read the manuscript entitled “Anomaly Detection Utilizing One-Class Classification – A Machine Learning Approach for the Analysis of Plant Fast Fluorescence Kinetics” with interest. This study established a one-class classification model using OJIP data exclusively from healthy lettuce (Lactuca sativa Larissa) plants. This model was then employed to predict the photosynthetic status of plants throughout their cultivation, classifying them as either “normal” or “anomaly”. Therefore, the manuscript needs some adjustments so that it can then be forwarded to the publication process. The manuscript has the potential for publication in the journal Stresses and needs the following adjustments:
TITLE
The authors analyzed only lettuce plants. Why would this study apply to plants in general? This question occurs throughout the manuscript. Review.
ABSTRACT
- JIP test was mentioned and in another section OJIP. Which would be correct? Standardize throughout the text.
- Check the use of “:” in objectives. I believe it is not necessary.
- Explore the results found further.
- Only one line was covered about the results.
- Add results and the conclusion. Expand the Abstract.
- Replace the keywords that are repeated in the title.
INTRODUCTION
- The first paragraph should be about the importance of fluorescence analysis in plants under stress conditions.
- It is not necessary to inform that PCA is a widely used analysis for studies with OJIP. This is relative, some researchers prefer it, others do not. This is not a rule. I suggest removing it.
- Why was Figure 1 mentioned in the Introduction? This is not correct. Move it to another section. The Figure was not cited in the text.
- In lines 62-67, it is necessary to insert a reference.
- Figure 2 was also mentioned here. This is not correct. Is the figure authored by the authors or from an external source? It was not cited in the text.
- In the objectives, which plants (species) were analyzed? Was the approach general or specific to a group of species?
- As mentioned, the study was conducted on lettuce. Why would the approach be for plants in general? Is this correct?
- Add hypotheses before the objectives.
RESULTS AND DISCUSSION
- In Figures 5 and 8, an axis with the values ​​within each graph was missing. This value should be normalized and not the raw values ​​used. See examples in the articles with chlorophyll a fluorescence kinetics.
- In Figures 8 and 9, this was not done. No statistical test to compare means?
MATERIAL AND METHODS
Although the data are from a previous study, information about its location is interesting (insert geographic coordinates).
- Was any statistical test used to compare treatments?
CONCLUSION
- Why Wasn't a thread added with Conclusions? This was not seen.
Author Response
We are grateful of the constructive comments from all reviewers and would like to address them point by point.
Comment: Dear Editors and Authors, I read the manuscript entitled “Anomaly Detection Utilizing One-Class Classification – A Machine Learning Approach for the Analysis of Plant Fast Fluorescence Kinetics” with interest. This study established a one-class classification model using OJIP data exclusively from healthy lettuce (Lactuca sativa Larissa) plants. This model was then employed to predict the photosynthetic status of plants throughout their cultivation, classifying them as either “normal” or “anomaly”. Therefore, the manuscript needs some adjustments so that it can then be forwarded to the publication process. The manuscript has the potential for publication in the journal Stresses and needs the following adjustments:
TITLE
The authors analyzed only lettuce plants. Why would this study apply to plants in general? This question occurs throughout the manuscript. Review.
Answer: Analysis of fast fluorescence kinetics is the widely used approach to study stress in plants and is not limited to a single species. Our study used data from lettuce, but it should be equally applicable to data from other plants.
Comment: ABSTRACT
- JIP test was mentioned and in another section OJIP. Which would be correct? Standardize throughout the text.
Answer: The kinetics of the rapid increase in chlorophyll fluorescence is referred to as the OJIP curve (or OJIP kinetics) because it consists of four distinct phases: O, J, I, and P. Thus, we have standardized the nomenclature: OJIP kinetics, OJIP curve, OJIP data, OJIP measurements, etc.
The only exception is the so-called JIP test, first developed by Strasser in 1995 to analyze the OJIP curve and transform it into a set of numerical parameters. Most of the parameters derived from the JIP test are calculated from the J, I, and P phases, hence the name. Since the name JIP-test is widely used in the literature, we decided to keep it.
Comment: Check the use of “:” in objectives. I believe it is not necessary.
Answer: We agree
Comment:
- Explore the results found further.
- Only one line was covered about the results.
- Add results and the conclusion. Expand the Abstract.
Answer: We believe that the abstract is an adequate summary of the key points of our study.
Comment: Replace the keywords that are repeated in the title.
Answer: We agree
Comment: INTRODUCTION
- The first paragraph should be about the importance of fluorescence analysis in plants under stress conditions.
Answer: We respectfully offer a different perspective on the matter. While fluorescence analysis in plants is indeed an important tool to study stress in plants, it is worth noting that this approach has been perfected in the last decades. In the Introduction we wanted to emphasize that there is still room for improvement in the use of artificial intelligence and machine learning to analyze chlorophyll fluorescence data. Thus, we have structured our discussion as follows: first, we describe the OJIP curve, then we discuss its widespread use for stress characterization, then we present the initial attempts to employ AI/ML and the still persisting problems, and finally, we conclude with the description of our novel approach.
Comment: It is not necessary to inform that PCA is a widely used analysis for studies with OJIP. This is relative, some researchers prefer it, others do not. This is not a rule. I suggest removing it.
Answer: It was not our intention to imply that PCA is a widely used analysis for studies with OJIP (nor did we explicitly state this in the text). The choice of whether or not to use PCA is, of course, at the discretion of the researchers. Our intention was merely to provide several examples of PCA used for OJIP analysis.
Comment: Why was Figure 1 mentioned in the Introduction? This is not correct. Move it to another section. The Figure was not cited in the text.
Answer: Figure 1 is an unpublished result by the author of this study. It illustrates the inhomogeneity of chlorophyll fluorescence, especially under stress conditions. Figure 1 was mentioned in the text (line 92). A line has been added to the legend of Figure 1 to reflect this fact.
Comment: In lines 62-67, it is necessary to insert a reference.
Answer: We agree
Comment: Figure 2 was also mentioned here. This is not correct. Is the figure authored by the authors or from an external source? It was not cited in the text.
Answer: Figure 2 was created by the author of the study. A line has been added to its legend to reflect this fact.
Comment: In the objectives, which plants (species) were analyzed? Was the approach general or specific to a group of species? As mentioned, the study was conducted on lettuce. Why would the approach be for plants in general? Is this correct?
Answer: As already mentioned, our study used data from lettuce, but it should be equally applicable to data from other plants.
Comment: Add hypotheses before the objectives.
Answer: Rather than testing hypotheses, we set out to determine the feasibility of our novel approach.
Comment: RESULTS AND DISCUSSION
- In Figures 5 and 8, an axis with the values ​​within each graph was missing. This value should be normalized and not the raw values ​​used. See examples in the articles with chlorophyll a fluorescence kinetics.
Answer: In Figures 5 and 8, the values are normalized with values from "normal" samples set to 1. In Figure 5, the small concentric circles on the right (with the red numbers) show the scales. We added the scale to Figure 8. We avoided adding individual scales to each plot, as they were already quite congested. A line of text is added to the legends to clarify this matter.
Comment: In Figures 8 and 9, this was not done. No statistical test to compare means?
Answer: We added statistical test to compare means.
Comment: MATERIAL AND METHODS
Although the data are from a previous study, information about its location is interesting (insert geographic coordinates).
Answer: We agree.
Comment: Was any statistical test used to compare treatments?
Answer: We added description of statistical test used in this study.
Comment: CONCLUSION
- Why Wasn't a thread added with Conclusions? This was not seen.
Answer: There was a Conclusion in the earlier version of the manuscript but it have been removed. Please ignore this.
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsDear author,
Thank you for your answers on the comments, they clarified several issues. I can recommend the manuscript to "accept".
Author Response
I would like to thank you for your time and comments.
Reviewer 2 Report
Comments and Suggestions for AuthorsI suggest removing the figures from the Introduction and moving them to another topic in the article, if necessary. Figures in the Introduction of research articles are not common. Despite being a previous study by the authors, the recommendation is to remove them.
Author Response
I have removed both figures in the Introduction. I would like to express my gratitude for your time and comments.
