AI-Based Virtual Assistant for Solar Radiation Prediction and Improvement of Sustainable Energy Systems

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper proposes and implements an AI-based hybrid prediction model integrating Random Forest, CatBoost, and Gradient Boosting, and incorporates this model into a Telegram chatbot. I believe the paper requires the following improvements:

1. The title does not adequately reflect the paper's innovation; revision is recommended.

2. The current literature review primarily lists existing methods without conducting a comprehensive overview or comparative analysis.

3. Data description lacks detail and omits fundamental statistical feature analysis.

4. Few comparative models are presented; incorporating common time-series comparison models is recommended.

5. The chatbot section should be the article's highlight, yet its description is severely limited. System architecture diagrams, interaction flowcharts, and similar structural elements should be added.

6. The paper claims greater flexibility compared to satellite systems like GLE Alert++ and HENON, but lacks empirical data or case studies to substantiate these advantages.

 

 

Author Response

We sincerely thank Reviewer 1 for the valuable comments and constructive suggestions, which have significantly contributed to improving the quality of our manuscript. Below, we provide a point-by-point response to each observation:

1. Improvement of the title
   Comment: The title does not adequately reflect the innovation of the article; a revision is recommended.
   Response: We have revised the title to better reflect the novelty of our work. The new title is:
   “AI-based virtual assistant for solar radiation prediction and improvement of sustainable energy systems.”

2. Literature review
   Comment: The current literature review mainly lists existing methods without providing a comprehensive overview or comparative analysis.
   Response: We expanded the literature review to include a more thorough and comparative analysis. In particular, we incorporated a detailed discussion of both the machine learning models applied in this study and the existing satellite-based models. Additional references have also been included to strengthen these comparisons in the introduction section.

3. Data description and statistical characteristics
   Comment: The description of the data lacks detail and omits the analysis of fundamental statistical characteristics.
   Response: We improved the data description by conducting a descriptive statistical analysis. The mean, median, maximum, and minimum values were evaluated to determine the representativeness and variability of the solar radiation records. This analysis revealed a balanced distribution of the dataset and is now explicitly presented in the data processing section.

4. Comparative models
   Comment: Few comparative models are presented; it is recommended to incorporate common time series models.
   Response: We have included an additional time series model along with its corresponding analysis. These results have been added in the penultimate row of Table 1, providing further comparisons with the existing models.

5. Expansion of the chatbot section
   Comment: The chatbot section should be the highlight of the article, but its description is very limited. It is recommended to include system architecture diagrams, interaction flow diagrams, and similar structural elements.
   Response: We fully agree with this observation. The chatbot section has been expanded in the revised manuscript. Additionally, system architecture and interaction flow diagrams have been incorporated to clearly illustrate the design, functionality, and interaction process of the proposed virtual assistant.

6. Flexibility compared to satellite systems
   Comment: The paper claims greater flexibility compared to satellite systems such as GLE Alert++ and HENON, but lacks empirical data or case studies to substantiate these advantages.
   Response: To address this observation, we added supporting bibliographic references in the introduction that confirm the greater flexibility of our approach compared to satellite systems such as GLE Alert++ and HENON. These references provide further support for the advantages highlighted in our study.

Reviewer 2 Report

Comments and Suggestions for Authors

Manuscript ID: sustainability-3885840

Title: Application of Artificial Intelligence to Predict Solar Radiation
Authors: Thomas Gavilánez, Néstor Zamora, Josué Navarrete, Nino Vega, Gabriela Vergara

General Evaluation:

The manuscript proposes a hybrid artificial intelligence model that integrates Random Forest, CatBoost, and Gradient Boosting to predict solar radiation levels, combined with a chatbot interface for real-time user alerts. The study is timely, with relevance to both renewable energy forecasting and public health protection from excessive UV exposure. The integration of AI models with a chatbot is innovative and potentially impactful for both environmental monitoring and user awareness.

The strengths of the paper include a comprehensive dataset, the implementation of multiple AI models, and the development of a practical chatbot application. The figures are clear and the comparison across methods is well presented. However, the paper is somewhat lengthy and occasionally repetitive, particularly in the description of the algorithms. Some methodological details (e.g., rationale for hybrid model design, hyperparameter optimization process, and validation strategy) need clearer justification. The limitations are acknowledged but could be more critically discussed, especially regarding data availability, temporal resolution, and the absence of cloud cover data. The user survey is a nice addition, though its scope and representativeness are limited. Overall, the paper is promising but requires “Major Revision” to improve clarity, critical analysis, and contextual framing.

Comments:

1. The abstract is informative but could emphasize more clearly what the chatbot adds beyond existing solar prediction systems (e.g., why is a chatbot more practical than a web dashboard or mobile app?).
2. The introduction cites a broad range of studies, but the novelty claim could be sharpened: what gap in current solar prediction and alert systems does this paper uniquely fill?
3. The methodology section contains extensive algorithmic detail; could you streamline the presentation by focusing on the unique aspects of the hybrid model and placing more standard algorithm equations in an appendix?
4. The dataset covers only 9 months from one station in Ecuador. How generalizable are the results? Would the model perform similarly with multi-year or multi-site data?
5. The hybrid model design (RF, GB, CB) is reasonable, but could you explain more clearly why these three were chosen and how their outputs are combined?
6. The validation metrics (MAE, MSE, R², F1-score) are appropriate, but it would be useful to include confidence intervals or error bars to show robustness of results.
7. I noticed that the number of Figure 18 is repeated, normally the figure in Page 22 must be Figure 19.
8. Also, in Figure 18 the left sub-figure, the language text is not written in English (I think we must present all captions or text n English)…Consider changing the language to English.
9. The discussion mentions limitations with timestamp homogenization and missing cloud cover data. Could you expand this into a stronger critique—how much do these factors impact the chatbot’s reliability in real-world use?
10. The user survey is interesting, but with only 20 participants it seems too limited to draw strong conclusions. Could you frame it more as a pilot usability test rather than evidence of adoption?
11. Figures are generally clear, but some (e.g., correlation plots, error curves) are crowded. Enlarging axis labels or simplifying presentation would improve readability.
12. The results show strong predictive performance, but the claim of 95% accuracy should be framed carefully—does this mean accuracy in classification of radiation levels or in regression performance?
13. Could you comment on computational cost—can the chatbot run predictions in real-time on modest hardware, or does it require substantial processing power?
14. The conclusion is informative but very descriptive; it could be reframed in simpler terms for broader readership, highlighting practical implications for public health, education, and renewable energy planning.
15. The reference list is broad, but you may want to include more recent works on AI-based real-time radiation prediction systems, especially studies that also include user interfaces.

Recommendation: Major Revision

 

Comments for author File: Comments.pdf

Author Response

We sincerely thank Reviewer 2 for the detailed and constructive comments, which have greatly helped improve the clarity, critical analysis, and contextual framework of our manuscript. Below, we provide a point-by-point response:

1. Abstract
   Comment: The abstract could more clearly emphasize what the chatbot adds beyond existing solar prediction systems.
   Response: We revised the abstract to highlight the advantages this research provides both for public health and for renewable energy generation and energy sustainability.

2. Introduction and novelty of the study
   Comment: The novelty claim could be sharpened: what gap in current solar prediction and alert systems does this paper uniquely fill?
   Response: We analyzed the gaps addressed by this research compared to previous studies and revised the introduction to clearly show these contributions, including additional recent references to support the work.

3. Methodology
   Comment: The methodology section contains extensive algorithmic details; it could be simplified by focusing on the unique aspects of the hybrid model.
   Response: The methodology was revised to focus on the hybrid model used for solar radiation prediction and the development of the chatbot, while standard algorithms were included in an appendix when appropriate.

4. Data generalizability
   Comment: The dataset covers only 9 months from a single station in Ecuador; how generalizable are the results?
   Response: In the final discussion, a paragraph was added acknowledging this limitation while highlighting the value of this research as a basis for future studies across longer periods or different locations.

5. Hybrid model design justification
   Comment: Could you more clearly explain why Random Forest, Gradient Boosting, and CatBoost were chosen and how their results are combined?
   Response: A paragraph was added detailing the specific advantages of each model within the hybrid scheme and how their combination improves solar radiation prediction.

6. Validation metrics
   Comment: It would be useful to include confidence intervals or error bars to show the robustness of results.
   Response: The results section now presents the performance of all models analyzed, emphasizing the benefits of the hybrid model compared to individual models.

7. Figure numbering
   Comment: Figure 18 is repeated; the figure on page 22 should be Figure 19.
   Response: All figures and their placement have been corrected, including those that were previously crowded.

8. Figure language
   Comment: Some figure text is not in English.
   Response: All figure text previously in Spanish was translated into English.

9. Data homogenization and cloud coverage limitations
   Comment: Could this be expanded with a more critical discussion?
   Response: We explain that through resampling and spline interpolation, data frequency was increased from 1 to 10 minutes, limiting the model’s ability to capture rapid radiation variations. Multi-frequency functions were used to maintain multiple resolutions in parallel, balancing robustness and sensitivity. Future research should consider higher-resolution time series using ideal sensors or advanced data fusion techniques.

10. User survey
    Comment: With only 20 participants, the survey is limited.
    Response: The survey was conducted as a usability pilot to validate prediction efficiency in a specific context.

11. Figure clarity
    Comment: Some graphs are overcrowded.
    Response: All crowded figures were improved for readability, adjusting labels and presentation.

12. Interpretation of 95% accuracy
    Comment: Does this refer to classification or regression?
    Response: We clarify that 95% indicates the model’s closeness to observed data, not overall solar prediction efficiency.

13. Computational cost and chatbot performance
    Comment: Can the chatbot perform real-time predictions on modest hardware?
    Response: Methodology now states that tree models (Random Forest, CatBoost) were chosen for both accuracy and low computational complexity, ensuring real-time predictions. Results indicate that the chatbot takes 2–3 seconds per prediction.

14. Conclusion
    Comment: Could be simplified for a broader audience.
    Response: The conclusion emphasizes practical implications for public health, education, and energy planning, highlighting how AI supports sustainability and renewable energy integration.

15. References update
    Comment: Include recent works on real-time AI-based solar prediction systems, especially those with user interfaces.
    Response: References were updated to include relevant works from the last 5 years that also involve user interfaces.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have done an excellent job, and the manuscript is well written. While there are already several UV prediction systems and applications available globally—such as the SunSmart Global UV app (a collaboration between the World Health Organization, the World Meteorological Organization, and other partners that provides real-time, location-specific UV forecasts and health guidance)—the development of a chatbot is a great idea.

That said, a few aspects could be strengthened. The survey sample size is currently limited to 20 participants, which is relatively small; expanding it to at least 50 participants would provide more robust insights. Additionally, the survey population could be extended beyond academia to ensure broader applicability and generalizability.

Overall, the proposed framework and tool are well designed, and I consider the paper suitable for publication without the need for major revisions.

1. One typo line 379 (usage of twice "API").
2. Improvement of figure quality such as Figure 9, 14 & 18.

 

Author Response

We sincerely thank Reviewer 3 for the positive and constructive comments, as well as for recognizing the relevance and innovation of our work. Below, we provide our responses to the observations:

1. Typographical error
   Comment: A typo was detected on line 379 (use of “API” twice).
   Response: The typographical error has been corrected in the manuscript.

2. Figure quality
   Comment: It is suggested to improve the quality of figures such as 9, 14, and 18.
   Response: The quality of all the mentioned figures has been enhanced to ensure clearer and more professional visualization.

3. Survey sample size and representativeness
   Comment: The survey sample is limited to 20 participants and could be expanded to ensure greater applicability.
   Response: It has been clarified in the manuscript that the survey was conducted as a usability pilot. The limitation of the sample size is acknowledged, and it is suggested that future research expand the study population to include more participants and a more diverse audience to improve the generalizability of the results.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author has revised the manuscript according to my comments and I recommend acceptance.

Reviewer 2 Report

Comments and Suggestions for Authors

All required comments have been addressed clearly. I suggest accepting the paper in its current form