An Interpretable Machine Learning-Based Framework for CO₂ Emission Prediction and Optimization: A Case Study of a University Campus

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Title: An Interpretable Machine Learning-Based Framework for CO₂ Emission Prediction and Optimization: A Case Study of a University Campus

Using data from a university located in Shandong Province, the study conducted carbon accounting and derived actionable plans under two scenarios—high-comfort soft control and low-comfort hard control. The results demonstrate strong applicability to campus "large-scale community" settings, enabling differentiated control across building types and seasons, and providing methodological support for regional CO₂ reduction strategies and the achievement of carbon-neutrality goals. The study is interesting and provides useful data about the CO₂ Emission Prediction and Optimization at a University campus. The manuscript can be published after revisions.

1. Adding references for the emission factor in Equation (1), (2), (3), (4), and (7).
2. Adding references for the carbon sequestration per unit area per day in Equation (5) and (6).
3. The Results and discussions section should be divided into the Results section and Discussion section.
4. Deleting Table 1 and explaining the parameter in text.
5. All variables (e.g., B) in Equation (10) should be explained.
6. Checking Equation (11).
7. Adding a Figure to show the annual total CO₂ emissions of different units at the university campus.
8. The language should be polished.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors Interpreting, analyzing, and forecasting carbon dioxide emissions is crucial for developing approaches to mitigating the impact of human activity on the environment. CO2 emissions are a multifaceted process, encompassing a number of interrelated parameters determined by a combination of pollutant sources, seasonal fluctuations, urban parameters, population number, and other factors. From a scientific perspective, machine learning-based models have recently become widely used to account for all these interrelated parameters and to accurately forecast and implement measures to mitigate greenhouse gas impacts. Despite this, a number of questions regarding the specifics of how to integrate building types, seasonal fluctuations, and time intervals into the model remain open. The present work addressing a number of such issues at univ campus level fills this gap. Using RF and SHAP modelling with detail account of variety of above-noted parameters, the authors develop self-consistent system of forecasting and optimization of carbon dioxide emissions. Although this was done for the specific case of a university campus, the model can be applied to other agglomerations with appropriate consideration of the input parameters.  Despite the fairly good study of the issue, the validity and consistency of the results, there is, nevertheless, the following comment to address 1)    From the description of the method and the relationships given on page 3, it is not entirely clear what significance the emission of other gases (for example, CO, NO, etc.) has (or does not have) and whether this needs to be taken into account, for example, when considering electricity consumption, transport activity, food consumption, etc.  Is it necessary to take into account the energy partitioning and divide the total energy consumed into the pollutants apart of CO2? This point needs clarification. In general, the manuscript is scientifically sound with the appropriate design to address the issues under consideration. It is clear, easy to follow, relevant for the field and presented in a well-structured manner. The manuscript provides sufficient details that support the conclusions. Referencing is quite comprehensive and up-to-dated. The manuscript is suitable for publication in the journal with account of the above minor point.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript presents an integrated, interpretable machine-learning framework for predicting and optimizing CO₂ emissions at a university campus. The study effectively combines Random Forest (RF) modeling, SHAP-based interpretability, and optimization algorithms to achieve actionable emission-reduction strategies under different comfort scenarios. The approach addresses both prediction accuracy and operational decision-making, which is highly relevant in sustainability research.

For better readability, some comments should be considered.

• The abstract doesn’t include sufficient outcomes of the manuscript.
• The authors should declare the scientific gap covered by their work. They only presented their objectives.
• The manuscript contains many symbols which should be better summarized in nomenclature table. Some symbols were not declared. For example, what are ?
• RF does well (R² = 0.92), however we can't tell which model is better because there aren't any comparisons to other models like XGBoost, Gradient Boosting, or Deep Neural Networks.
• The RF model seems to have been trained on aggregated temporal data, like monthly or seasonal data, although the authors talk about hourly and daily features. We need to know more about how the model's ability to make predictions changes with different levels of temporal granularity.
• Scenario design (high- vs. low-comfort) lacks explicit quantitative metrics for “comfort.” How are temperature thresholds, indoor air quality, or occupant satisfaction quantified?
• The conclusion doesn’t reflect the efforts in preparing this manuscript. A polished version is required to explore the main outcomes.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Accept.