All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Thank you for your contribution.
[# PeerJ Staff Note - this decision was reviewed and approved by Xiangjie Kong, a PeerJ Section Editor covering this Section #]
The authors revised the manuscript adequately according to the reviewer comments.
The manuscript is now more qualified and clear.
I have no further comments.
I suggest accepting it for publication in its present form.
-
-
-
This version looks better, thanks for addressing the review comments.
This version looks better, thanks for addressing the review comments.
This version looks better, thanks for addressing the review comments.
This version looks better, thanks for addressing the review comments.
Thank you for the revision. Some responses are limited. Pay attention to the metrics section.
The authors revised the manuscript adequately according to the reviewer comments.
The manuscript is now more qualified and clear.
I have no further comments.
I suggest accepting it for publication in its present form.
Accept
Accept
This version looks better, thanks for addressing the review comments.
This version looks better, thanks for addressing the review comments.
This version looks better, thanks for addressing the review comments.
This version looks better, thanks for addressing the review comments.
All comments are for the last part.
All comments are for the last part.
All comments are for the last part.
Thank you for the revision. Some responses are limited. Pay attention to the metrics section.
Please follow the reviewers' requests and comments in detail.
**PeerJ Staff Note:** Please ensure that all review and editorial comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** The review process has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at [email protected] for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
This study aims to build a deep learning model to classify chart images, including Histogram, Bar, Line, and Pie charts.
It is a valuable study. However, the quality of the paper can be improved as follows:
1- The main contributions of the study are not clear.
The novelty (main contributions) of the study may be defined in the "Introduction" section clearly, i.e., in a single separate paragraph and point-by-point.
2- What is the running time (execution time) of the methods?
Additional results may also be given in terms of execution times (computational cost).
3- In addition to accuracy and confusion matrix, it can be useful if the results of other metrics (e.g., precision, recall, F-measure) are also given.
4- The authors may explain the possible future studies in the conclusion section.
5- The authors should provide the full forms of the abbreviations upon their first use.
Furthermore, some abbreviations are used in the paper without being defined, such as ADC, CCPR, mAP, RMSProp, and ResNet.
6- Figures 4 and 5 should also show "x-axis title" and "y-axis title".
7- A concern is that no formal statistical analysis of the results is done to indicate whether the differences in performance are statistically significant or not.
For example, Friedman Aligned Rank Test, Wilcoxon Test, Quade Test, etc.
P-value can be calculated and compared with the significance level (p-value < 0.05).
8- Table 2 presents the parameters for all selected models. However, some values of some parameters are missing, such as "learning rate".
-
-
Language is often verbose and unpolished. Needs proofreading.
The introduction covers too many ML basics not relevant to the core focus.
Figures and tables are present, but explanations need improvement.
Dataset and model setup are valid and within journal scope.
Custom CNN is a reasonable contribution.
However, details on hyperparameters, reproducibility, and evaluation metrics are scattered.
Classification accuracy is promising.
Consider adding statistical or error-based validation.
Lacks generalizability and discussion on limitations.
This article addresses an emerging area in visual document understanding — classifying chart images — with an emphasis on unsupervised learning. Thanks for the nice article.
Abstract:
1. The abstract includes long, grammatically incorrect sentences that reduce readability.
2. While the abstract mentions the methods and results, it fails to state the specific research question or objective of the study.
3. Terms like “RESTNET50” instead of “ResNet50” and “financial, social, economic” (should be “socio-economic”) must be corrected. Please use consistent terminology.
Introduction:
1. The introduction broadly discusses the challenges in character recognition but does not clearly define the gap this article is addressing. Can you include a clear statement?
2. Several paragraphs in the Introduction delve into algorithm-level detail and general ML techniques that are better suited for the Related Work section. Streamline the introduction to highlight why chart classification is important and why unsupervised methods matter.
3. The explanation of supervised, unsupervised, and reinforcement learning is too long for an Introduction section. Consolidate this into one paragraph or move to a separate “Background” section if needed.
Related Works:
1. The review mixes domains such as healthcare, finance, and geotechnical engineering without grouping them by methodology or relevance.
2. The article summarizes prior studies but does not contrast them effectively.
Materials and Methods:
1. The methodology flow is not described step-by-step. Authors should clearly outline data preprocessing steps (image size, format, normalization), Feature extraction (ex., which layers of VGG16), Clustering process (why 4 number of clusters?), and evaluation metrics used, etc.
2. The text does not explain why PCA is necessary before k-means or how dimensionality was reduced. Can you justify the use of PCA?
3. Though figures are included, they lack detailed architectural design for the Custom CNN. You may need to include a flow diagram or architecture schematic showing all layers and parameters.
Experimental Setup:
1. You may need to add missing hardware /software specifications for reproducibility.
2. You may have to disclose hyperparameter training, ex. epochs, batch size, learning rate, optimizer used, and dropout rates must be clearly mentioned.
3. Can you justify the split of 60/20/20 for train/val/test? Can you include a brief rationale and verify if the class distribution was balanced across splits?
Results and Discussion:
1. The discussion does not explore why models like DenseNet fail on pie charts. You need to perform error analysis or misclassification reasoning.
2. Accuracy alone is insufficient. You may need to include precision, recall, F1-score, and ideally, cross-validation or confidence intervals.
3. The article lacks an ablation or comparative study to highlight why Custom CNN outperforms other models.
Conclusion:
1. The conclusion does not include any numerical performance metrics to support claims of effectiveness. Consider adding key quantitative results.
2. The conclusion makes broad claims like “transform different fields” without direct evidence.
3. Every conclusion should include limitations. You may need to add that.
4. For future directions, add suggestions such as exploring more diverse chart types, synthetic data augmentation, etc.
5. The conclusion refers to the pipeline as an “unsupervised classification system,” in several components, in my observation, such as CNN feature extraction and supervised classifier training, which involve labeled data. Can you clarify or reframe the language?
All comments have been added in detail to the last section.
-
-
1. In this study, high accuracy was achieved in classifying chart images using VGG16 for feature extraction, PCA for signal extraction, k-means clustering, and pre-trained models such as RESTNET50, RESTNET50 V2, DenseNet, and a customized CNN for visualizing financial, social, economic, and political data with chart types like Histogram, Bar, Line, and Pie.
2. In the introduction section, visualization through chart images, Data-Driven decisions, the importance of the subject, and machine learning-related studies on the subject were mentioned at a certain level. The information and emphasis of the subject in this section are appropriate and sufficient.
3. In the related works section, studies on machine learning, deep learning, and hybrid approaches in the literature related to the classification of charts and graphs were mentioned. In this section, the literature table should be detailed and referenced to table-1 should be given.
4. It was stated that the ChartVQA dataset was preferred as the dataset in the study. This dataset is at a suitable level for this study in terms of both quantity and data types. However, it is still useful to detail why this dataset is preferred compared to other datasets in the literature.
5. When the data preprocessing steps are examined, it is observed that it is at a very basic level, but it is understood that these preprocessing steps are at a sufficient level before classification for the study.
6. It is stated that VGG and K-means are used for feature extraction and clustering, respectively. In this section, although there are many deep learning-based structures that can be used instead of VGG, it should be detailed with the reasons why this is preferred.
7. There are very serious deficiencies in evaluation metrics. Obtain all the missing metrics, especially the receiver operating characteristic (ROC) curve and AUC (the area under the ROC curve) score.
As a result, for the study to make a full contribution to the literature, the above items must be addressed completely.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.