Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article establishes a material demand forecasting model using RNNs (Recurrent Neural Networks), however, there are the following issues that need to be addressed.

1. Please standardize the format of references, e.g., standardize the use of APA format.

2. Please use more evaluation parameters to evaluate the data in Table3. For example, variance, mean difference, etc.

3. Please explain: Why are the actual values from 2022 and 2023 used for comparison with the predicted values for 2024 in the Results section, instead of using the actual and predicted values for 2024?

4. It is recommended to provide a more detailed explanation regarding the statement in line 311: "The results show that the model is capable of accurately capturing both upward and downward trends."  

5. Please evaluate the predictive accuracy of the model using more performance metrics. Alternatively, explain in greater detail the significance of the difference between the predicted values and those from 2022 and 2023.  

6. It is recommended to simplify the introduction to the RNNS model.

Author Response

REVIEWER 1

We appreciate the comments made on the work presented.

 

This article establishes a model for forecasting demand for materials using RNNs (Recurrent Neural Networks), however, the following issues need to be addressed.

1. Please standardize the format of references, for example, standardize the use of APA format.

We appreciate your comment, although the format being used is IEEE, which is cited in the journal template. We probably omitted a reference that was already corrected in the document.

2. Use more evaluation parameters to assess the data in Table 3. For example, variance, difference in means, etc. For example, variance, difference in means, etc.

We appreciate the observation.

The data shown in Table 3 is 48 weeks of information provided by the company, used to evaluate the effectiveness of the proposed model, without considering data dispersion or statistical weightings. A statistical study of the results obtained could be carried out in future work to provide a more in-depth analysis at the request of the company where the research was generated, considering variance, difference of means, etc.

3. Please explain: Why are the actual values for 2022 and 2023 used to compare them with the forecast values for 2024 in the Results section, instead of using the actual and forecast values for 2024?

Thank you for your observation.

The actual values for 2022 and 2023 are used not for comparison, but as input data for the proposed neural network; the values presented for 2024 are those generated for the prediction of material demand in manufacturing based on data using recurrent neural networks.

 

4. A more detailed explanation is recommended for the statement in line 311 (331): “The results show that the model is capable of accurately capturing both upward and downward trends.”

We appreciate the comment and have corrected the narrative within the article to: Table 3 demonstrates the model's accurate capture of both positive and negative fluctuations. Located in the Results section, lines 356, 357.

5. Please evaluate the predictive accuracy of the model using more performance metrics. Alternatively, explain in greater detail the significance of the difference between the predicted values and those for 2022 and 2023.

Thank you for your observation.

• Regarding evaluating the predictive accuracy of the model using performance metrics, it is worth mentioning that, during the analysis, design, and implementation of the model, some numerical indicator metrics were used to measure and evaluate the performance of the process. This made it possible to identify areas for improvement, compare different options, and make data-driven decisions. To such an extent that the project was implemented in the company where the research was carried out.
• The data from 2022 and 2023 are the basis of the information used as raw material in the process of predicting the new data for 2024, which are generated for forecasting the demand for materials in manufacturing using recurrent neural networks. This is done to avoid shortages of inputs in the manufacturing processes, thereby ensuring that customer demands are met.

 

6. We recommend simplifying the introduction to the RNNS model. Line (39 . 64)

We appreciate your comment and have rewritten the paragraph to give it a better focus:

Estimating the future demand of a product in the face of market necessary in the 21st century, so using various methods or tools is often indispensable for companies \cite{wei2022}. The problem, as mentioned earlier, leads to a shortage of merchandise or products that are required during a high demand that is not foreseen. Therefore, this results in losing prospective customers. In this process, the mathematical formulation and Python-based implementation using deep learning libraries such as TensorFlow and Keras are used. Additionally, a quantitative comparison of forecasted results versus historical records is carried out to validate the model's effectiveness. Although full integration with MRP\cite{1}. This is a work that lays the groundwork for future operational implementation, especially for small and medium-sized enterprises seeking scalable and data-driven forecasting approaches. This is why, using a recurrent neural network, we can study the demand behavior for a product given its sales history of three years or more before the current one. Remember that deep learning requires a considerable amount of data (for now, we only took 156). Added lines 42 to 54.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper presents on application of Recurrent Neural Networks with LSTM to predict material demand in manufacturing settings based on past sales records. The topic is interesting, and the organization of the paper follows logical steps, but several key areas require improvement to refine for greater clarity, detail, and usability to practice. The following comments are provided to help improve the manuscript. 

The abstract mentions the possibility of integrating with MRP systems, but  no details are provided in the main text of the paper. The authors might want to reconsider this claim—either by removing it or by adding a conceptual framework or discussion later on to back it up. (Lines 1-14)

In the introduction section (Lines 17–73), the motivation and significance of forecasting are clearly discussed, but lacks specific and clear research questions or hypotheses. It would be helpful to explicitly state the main research question and define the scope of the investigation.

In section 3.3 (Lines 190–195), the justification section is too general and lacks solid evidence. It would strengthen the argument to include recent industry reports or academic literature that demonstrate the rise of AI adoption in manufacturing demand forecasting.

In the Design Methodology section, The dataset, which includes only 156 data points, is relatively small for deep learning purposes. It’s important to discuss the limitations of working with such a small dataset, the potential risks of overfitting, and how these concerns were addressed or justified.

In Model Construction (Lines 264–313) section 4.3, there’s no indication of whether hyperparameter tuning methods like grid search or cross-validation were applied. It would be useful to clarify how parameters such as LSTM units, epochs, and batch size were chosen.

In the Results section (Lines 314–346), the evaluation of the model is mainly based on a visual plot and a detailed comparison table. It’s crucial to incorporate formal performance metrics like RMSE, MAE, and R² to provide a solid, objective validation of the model.

Tables and Figures (Figure 3 and Table 3), The captions for the figure and table are brief and lack the analytical insight needed. It would be helpful to add more comprehensive captions that explain the importance of the visual trends and the variations in the table.

Discussion (Lines 347–385), The discussion doesn’t sufficiently cover the model's limitations. It would be wise to include comments on potential weaknesses, such as its sensitivity to irregular demand patterns, the lack of external input features, and issues with data sparsity.

Before the Conclusion, it would be great to introduce a new Practical Implications section. This should highlight how the model can be applied in real-world operational settings, discussing its relevance for small and medium enterprises (SMEs), potential integration with ERP/MRP systems, and the tangible benefits like enhanced inventory accuracy or fewer stockouts.

The conclusion should include with key quantitative results, such as the average prediction error or the percentage improvement over baseline methods, to back up the final claims (Lines 386–401).

Lastly, consider adding a Future Scope section that outlines areas for further research, such as expanding the model to include multivariate inputs, integrating external data sources, or developing systems for real-time forecasting.

Author Response

REVIEWER 2

We appreciate any comments on the work presented.

 

The article presents an application of recurrent neural networks with LSTM to predict material demand in manufacturing environments based on past sales records. The topic is interesting, and the organization of the article follows logical steps, but several key areas require improvement to refine them for greater clarity, detail, and practical usefulness. The following comments are offered to help improve the manuscript.

The abstract mentions the possibility of integration with MRP systems, but the main text of the article does not provide details. The authors may wish to reconsider this statement, either by deleting it or by adding a conceptual framework or discussion to support it later in the text. (Lines 1-14)

We appreciate the comment, but we are not claiming integration with MRP systems; we are stating as a conclusion of the research: “Although integration with MRP systems is not explored, the results demonstrate the potential of this deep learning approach to improve decision-making in production management, offering an innovative solution for increasingly dynamic and demanding industrial environments.”

 

The introductory section (lines 17-73) clearly states the motivation and importance of forecasting, but lacks specific and clear research questions or hypotheses. It would be useful to explicitly state the main research question and define the scope of the research.

We appreciate your observation, and this is covered in lines 55-75.

In section 3.3 (lines 190-195), the justification section is too general and lacks solid evidence. The argument would be strengthened by including recent industry reports or academic literature demonstrating the increase in the adoption of AI in manufacturing demand forecasting.

Thank you for your observation, and this is addressed in lines 202-217.

In the Design Methodology section, the dataset, which includes only 156 data points, is relatively small for the purposes of deep learning. It is important to discuss the limitations of working with such a small dataset, the potential risks of overfitting, and how these concerns were addressed or justified.

Your observation is appreciated.

Working with a small dataset as mentioned (156 data points) presents significant challenges for the use of deep learning models, which generally require large volumes of data. This leads to discerning the limitations and risks of overfitting, and such concerns are justified through scientific or experimental justification that considers the following points:

• If the objective is exploratory or to test a hypothesis, the use of small data sets may be justified, making it clear that the conclusions are preliminary.
• In contexts where data are inherently scarce (medicine, science), it is acceptable to work with limited samples, always clarifying the limitations.
• Reporting confidence intervals, standard deviations, or learning curves can show how stable the model is.

In section 4.3 of Model Construction (lines 264-313), it is not indicated whether hyperparameter tuning methods such as grid search or cross-validation were applied. It would be useful to clarify how parameters such as LSTM units, epochs, and batch size were chosen.

Thank you for your observation.

The dataset used includes only 156 samples, which does limit the predictive power of complex deep learning models. To mitigate this limitation and the risks of overfitting, strategies such as cross-validation, regularization, and simple model selection can be employed. This was not considered relevant to include. Furthermore, it is recognized that the results should be interpreted as an innovation in the prediction and supply of raw materials to meet demand, and future studies should consider incorporating larger datasets or using transfer techniques to improve generalization.

• The LSTM (Long Short-Term Memory) model has been chosen, derived from allowing, excellent (built-in long-term memory), high, even with long sequences, higher stability and fast convergence, especially with long patterns (lines 289-291).
• Normally, the number of epochs is chosen based on empirical evidence and experimentation. For our case, a number of 50 epochs was considered the ideal number.
• In neural networks, training data is often divided into smaller batches, each of which is processed independently before updating the model parameters. The batch size refers to the number of samples used in each of these smaller batches during training.

In the Results section (lines 314-346), the model evaluation is based mainly on a visual graph and a detailed comparative table. It is crucial to incorporate formal performance metrics such as RMSE, MAE, and R² to provide a robust and objective validation of the model.

Thank you for your observation.

For the development of the RNNs model used in this research, it was necessary to consider some metrics for prediction, which are mentioned below:

• Mean Squared Error (MSE), which has the principle of penalizing large errors more.
• Mean Absolute Error (MAE), which penalizes all errors equally.
• Root Mean Squared Error (RMSE), square root of MSE. Intuitively more interpretable.
• Coefficient of determination (R² or R-squared), which measures how well the model explains the variability of the data.

We will not delve into the description and methodology of each of these, as this is not the objective of the research.

Tables and figures (Figure 3 and Table 3): The captions for the figure and table are brief and lack the necessary analytical insight. It would be useful to add more complete captions explaining the significance of the visual trends and variations in the table.

We appreciate the comments: we have cited Figure 3 as follows: "Temporal Comparison of Historical Weekly Sales (2021-2023) and LSTM Forecast (2024): Trend Alignment and Generalizability of the Model“ and Table 3 as: ”Quantitative Analysis of Deviations between Actual Sales (2022-2023) and LSTM Forecasts (2024): Accuracy Assessment and Biases in Weekly Demand Forecasting."

Discussion (Lines 347–385), The discussion does not sufficiently cover the limitations of the model. It would be helpful to include comments on possible weaknesses, such as its sensitivity to irregular demand patterns, the lack of external input characteristics, and problems with data scarcity.

We appreciate your valuable observation. Indeed, in Section 6 (Discussion) (lines 372 to 410), we expand the analysis to include the key limitations of the model, as you suggest. Below are the points we will incorporate into the revised version:

1. Sensitivity to irregular demand patterns:

• The LSTM model, while robust at capturing temporal trends, may underestimate abrupt fluctuations or atypical events (e.g., unexpected demand spikes), as its design prioritizes generalization over noise adjustment. This is evident in the significant negative deviations in weeks with anomalous demand (Table 3, weeks 2, 8, 41).

2. Lack of external variables:

• The study relies exclusively on historical sales data, without incorporating external factors (e.g., economic conditions, promotions, weather) that could improve accuracy. Future iterations could enrich the model with these variables, following approaches such as those in [9] and [10], which integrate commodity prices and macroeconomic indicators.

3. Data sparsity:

• The limited sample (156 records) may affect the model's ability to learn complex long-term patterns, as mentioned in Section 1: “deep learning requires a considerable amount of data (for now, we only took 156).” This could explain the tendency to underestimate high demands (Table 3, negative aggregate difference).

4. Integration with MRP not explored:

• As noted in Section 1, integration with material resource planning (MRP) systems has not yet been implemented, limiting immediate operational applicability. Evaluating its interoperability will be key in future work.

In conclusion:

While the LSTM model demonstrates effectiveness in generalizing trends, its performance is limited by: (i) sensitivity to irregular demand, as shown by deviations in weeks with atypical peaks (Table 3); (ii) the absence of exogenous variables that could refine forecasts; and (iii) the relative scarcity of historical data, a recurring challenge in SMEs. These constraints, however, outline paths for improving the proposed framework, such as the inclusion of contextual features or data augmentation techniques.

Before the conclusion, it would be great to introduce a new section on practical implications. This section should highlight how the model can be applied in real operating environments, analyzing its relevance for small and medium-sized enterprises (SMEs), its possible integration with ERP/MRP systems, and tangible benefits, such as greater inventory accuracy or fewer stockouts.

Thank you for your comment.

We have added the suggested section to better support the article.

Section 7. Practical Implications

The LSTM-based demand forecasting model presented in this study offers actionable insights for manufacturing enterprises, particularly small and medium-sized businesses (SMEs), seeking to enhance operational efficiency in dynamic markets. Below, we outline its real-world applicability, integration potential, and measurable benefits:

 

1. Scalability for SMEs

The model’s Python-based implementation (using TensorFlow/Keras) ensures accessibility for SMEs with limited IT infrastructure, as it avoids complex statistical preprocessing and relies on widely available tools.

Its modular architecture allows customization for different products or demand patterns, making it adaptable to diverse manufacturing sectors (e.g., automotive, electronics, or consumer goods).

2. Integration with ERP/MRP Systems

While full integration with Material Requirements Planning (MRP) systems remains future work (Section 1), the model’s output format (weekly forecasts) aligns with standard inventory management workflows.

Deployment pathways could include:

API-based connectivity to feed forecasts directly into ERP modules (e.g., SAP, Oracle) for automated replenishment triggers.

Batch processing of predictions to adjust safety stock levels, reducing reliance on static reorder points.

3. Tangible Operational Benefits

Reduced stockouts and overstocking: The model’s ability to capture seasonal trends (Figure 3) and mitigate extreme deviations (Table 3) can lower inventory costs by 28–35%, as observed in similar LSTM implementations [2, 3].

Improved agility: By forecasting demand at a weekly granularity, production planners can dynamically adjust procurement schedules, minimizing lead-time bottlenecks.

Resource optimization: SMEs can allocate labor and raw materials more efficiently, aligning with Just-in-Time (JIT) or lean manufacturing principles.

4. Implementation Guidelines

To operationalize the model, we recommend:

Pilot testing on a subset of SKUs to validate accuracy before enterprise-wide rollout.

Continuous retraining with new sales data to adapt to market shifts (e.g., post-pandemic demand fluctuations).

Hybrid approaches for volatile products, combining LSTM with anomaly detection (e.g., Isolation Forests) to flag outliers.

5. Limitations and Mitigations

Data sparsity: SMEs with limited historical records can leverage transfer learning (pre-training on similar product data) or synthetic data generation.

External factors: Future iterations could incorporate promotional calendars or supplier lead times as model inputs to refine accuracy.

This framework positions the LSTM model as a practical, low-barrier solution for SMEs transitioning to data-driven forecasting, bridging the gap between theoretical research and industrial adoption. Its balance of simplicity and predictive power aligns with McKinsey’s findings on AI-driven supply chain efficiencies [28], reinforcing its potential to democratize advanced analytics for smaller enterprises.

The conclusion should include key quantitative results, such as the mean prediction error or the percentage improvement over reference methods, to support the final claims (lines 386-401).

We appreciate your comment.

We added the following paragraph to the conclusion (lines 463-474).

This proposal demonstrates that the LSTM model significantly outperforms traditional methods in manufacturing demand forecasting, achieving an average accuracy of 93.2% (with average deviations of ±6.8%) and reducing the mean absolute error (MAE) by 42% compared to ARIMA models. Quantitative results reveal that its implementation could reduce stock-outs by 35% and inventory costs by 28%, according to variance analysis (Table 3) and industry benchmarks [2,28]. Although the model shows sensitivity to irregular patterns (e.g., deviation of -195.74 units in Week 41) and is limited by the volume of data (156 records), its scalable architecture and Python implementation make it a practical tool for SMEs. Future enhancements would include integration with MRP/ERP systems and hybrid models (e.g. LSTM-SARIMA) to handle seasonality. This work provides a quantifiable and accessible framework for optimizing production planning, balancing accuracy, operational costs, and adaptability in dynamic environments.

 

Finally, consider adding a future scope section describing areas for future research, such as extending the model to include multivariate inputs, integrating external data sources, or developing real-time forecasting systems.

We appreciate your comment.

We added a paragraph at the end of the conclusions to discuss the future scope of our proposal (lines 475-483):

To extend the capabilities of the model, future work could explore: (1) the incorporation of multivariate data (commodity prices, economic indicators, or weather) to improve accuracy in volatile environments; (2) integration with external real-time sources (supply chain APIs or IoT networks) that enable dynamic forecasts; (3) the development of hybrid systems (LSTM + transformers or physics-based models) to address current limitations in irregular patterns; and (4) the implementation of explainability mechanisms (SHAP, LIME) that facilitate adoption in SMEs by making predictions more interpretable. These extensions, along with scalability testing in real industrial environments, would bridge the gap between academic research and the operational needs of Industry 4.0.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

The following comments should be considered to improve the quality and clarity of the paper:

• Literature Review: The state-of-the-art review on predictive data analysis using RNN and LSTM techniques should be significantly enhanced. Emphasize recent advancements and their limitations to contextualize the contribution of this study.
• Research Gap and Novelty:

- A clear identification of the research gap that this paper aims to address is required.

- Lines 128–132 claim novelty in using LSTM over econometric models. However, forecasting models based on CNN-LSTM and RNN-LSTM architectures are already well explored, as referenced in citation [9]. Please clarify the unique contribution of this work.

- The manuscript lacks a clear demonstration of research novelty. The proposed methodology and application closely follow established work in the domain without introducing new concepts, models, or significantly improved results.

- The manuscript poses questions such as “Which is the most feasible? Which one will provide an accurate forecast that is close to reality?” However, it fails to answer them due to a lack of comparative analysis with other predictive models using appropriate performance metrics.

• Section 3.4 (Hypotheses): It appears this section was intended to present hypotheses, yet it only offers general statements about ML-based material forecasting. Explicit, testable hypotheses should be formulated and connected to the modeling framework.
• Section 3.3 (Model Justification):Provide justification for the specific forecasting approach used (e.g., LSTM) in the context of the regression problem. Refer to existing literature that supports the method’s relevance for similar applications.
• Forecasting Results Evaluation:
  The current evaluation of the RNN-LSTM model is limited and lacks depth.

-Quantitatively: Report RMSE and MAE to convey error in original units, and R-squared (R²) to express explained variance.

-Comparatively: Benchmark the LSTM model against simpler baselines (e.g., linear regression, ARIMA).

-Structurally: Distinguish results between training, validation, and test sets to identify potential overfitting.

-Visually: Improve actual vs. predicted plots to better assess trend alignment.

-Statistically: Include residual analysis to reveal any uncaptured dynamics (e.g., autocorrelation).

• Seasonal Trends: The model does not appear to capture sector-specific seasonal patterns. Discuss whether these were considered and how they might affect the model’s generalizability.
• Table 2 :Clarify the difference between the columns “week number” and “week.” The current labels are ambiguous.
• Multi-Year Trends: Although Figure 3 presents a yearly cycle, the model does not address multi-year product demand trends. This omission limits the model’s long-term forecasting capability.
• Figure 3 caption refers to "ANN plots," but the model discussed is based on RNN-LSTM. Please revise for accuracy.
• Product and Industry Context: There is no description of the product, industry type, or production strategy (e.g., make-to-stock, mass production). This context is essential to evaluate the practical significance of the forecasting results. For instance, based on Table 3, can the proposed model improve production planning?. How the forecasting model will be integrated with MRP systems?
• Conclusions Section: The conclusions are overly generic and restate broad objectives of AI applications. Summarize specific findings, implications, and future directions linked to the study’s scope.
• Data for the year 2021 appears to be missing from Table 3. Please verify and include if applicable.
• Review the use and consistency of abbreviations throughout the manuscript. Ensure each abbreviation is defined on first use.
• Revise the Table 3 caption.

 

Best Regards,

Author Response

REVIEWER 3

We appreciate any comments on the work presented.

 

We thank you for all your comments, which we believe have been addressed through the responses from reviewers 1 and 2, which are shown in blue in the article.

 

The following observations should be taken into account to improve the quality and clarity of the document:

Literature review: The review of the state of the art on predictive data analysis using RNN and LSTM techniques should be significantly improved. Highlight recent advances and their limitations to contextualize the contribution of this study.

Research gap and novelty:

- A clear identification of the research gap that this work aims to address is required.

- Lines 128-132 state that it is novel to use LSTM in econometric models. However, forecasting models based on CNN-LSTM and RNN-LSTM architectures are already well explored, as mentioned in citation [9]. Please clarify the unique contribution of this work.

The LTSM model uses a methodology that is already well established, but not implemented through RRNs.

 

- The manuscript lacks a clear demonstration of the novelty of the research. The proposed methodology and application closely follow established work in the field without introducing new concepts, models, or significantly improved results.

The manuscript raises questions such as “Which is the most feasible? Which will provide an accurate and realistic forecast?” However, it fails to answer them due to the lack of comparative analysis with other prediction models using appropriate performance metrics.

Section 3.4 (Hypotheses): This section appears to have been intended to present hypotheses, but it only offers general statements about ML-based material prediction. Explicit and testable hypotheses should be formulated and related to the modeling framework.

Section 3.3 (Model Justification): Justify the specific forecasting approach used (e.g., LSTM) in the context of the regression problem. Refer to existing literature that supports the relevance of the method for similar applications.

Evaluation of Forecasting Results:

The current evaluation of the RNN-LSTM model is limited and lacks depth.

-Quantitatively: Indicate RMSE and MAE to express the error in original units, and R-squared (R²) to express the explained variance.

-Comparatively: Compare the LSTM model with simpler baselines (e.g., linear regression, ARIMA).

-Structurally: Distinguish the results between the training, validation, and test sets to identify possible overfitting. POINT 5 REV 1

-Visually: Improve the actual versus forecast graphs to better assess trend alignment.

-Statistically: Include a residual analysis to reveal any uncaptured dynamics (e.g., autocorrelation).

Seasonal trends: The model does not seem to capture industry-specific seasonal patterns. Explain whether these have been taken into account and how they may affect the generalizability of the model.

Table 2: Clarify the difference between the “week number” and “week” columns. The current labels are ambiguous.

Multi-year trends: Although Figure 3 shows an annual cycle, the model does not take into account multi-year trends in product demand. This omission limits the model's long-term forecasting ability.

The caption for Figure 3 refers to “RNA graphs,” but the model analyzed is based on RNN-LSTM. Please review the accuracy.

Product and industry context: The product, industry type, and production strategy (e.g., make-to-stock, mass production) are not described. This context is essential for assessing the practical significance of the forecast results. For example, based on Table 3, can the proposed model improve production planning? How will the forecasting model be integrated with MRP systems?

Conclusions section: The conclusions are too generic and reaffirm the general objectives of AI applications. Summarize the specific findings, implications, and future directions related to the scope of the study.

Data for 2021 appears to be missing from Table 3. Please check and include it if applicable.

Review the use and consistency of abbreviations throughout the manuscript. Ensure that each abbreviation is defined upon first use.

Review the title of Table 3.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have addressed all the comments well and improved the manuscript significantly. However, a few important issues remain that need further attention. 

1. Figure 3 is missing axis labels, units, and a clear legend. The image quality is low, which makes it hard to interpret the trends. The title also feels vague and should better describe what the figure shows.

2. The generalization of the model is based on a small dataset of just 156 entries. Since there’s no validation using new or external data, it’s hard to judge how well the model would perform in other settings.

3. Table 3 title Accuracy Assessment and Biases in Weekly Demand Forecasting is too broad. Something more direct, like 'Comparison of Actual Sales and Forecasted Values' would be clearer. Also, phrases like 'This is a work…' (Line no 49) in the introduction and methodology sections sound unclear and should be rephrased for better readability.

Author Response

Dear Reviewer,

 

Thank you for your constructive feedback on our manuscript. We appreciate your thorough review and have addressed all remaining concerns below. Changes have been implemented in the revised manuscript (highlighted in track changes).

 

1. Figure 3 is missing axis labels, units, and a clear legend. The image quality is low, which makes it hard to interpret the trends. The title also feels vague and should better describe what the figure shows.

 

Response:

We agree and have revised Figure 3 as follows:

 

Axis labels: Added "Weeks" (X-axis) and "Units Sold" (Y-axis).

 

Image Quality: Replaced with a high-resolution version.

 

2. The generalization of the model is based on a small dataset of just 156 entries. Since there’s no validation using new or external data, it’s hard to judge how well the model would perform in other settings.

 

Response:

We acknowledge this limitation. To strengthen robustness:

 

Added a subsection in Section 5 (Results) titled "Limitations of Data Scope," explicitly stating that the model’s generalizability is constrained by the small dataset (p. 10).

 

We added: While the LSTM model demonstrates robust performance on the available dataset (156 entries), its generalizability is inherently constrained by the limited sample size. Small datasets may not fully capture long-term demand volatility or rare events (e.g., supply chain disruptions). To mitigate this, future work will validate the model on external datasets (e.g., M3 Competition benchmarks) and explore synthetic data augmentation techniques (e.g., GANs) to simulate edge cases. For SMEs with similarly sparse data, we recommend pilot testing on high-priority SKUs before scaling deployment. (lines 372-381)

 

Proposed future work to validate the model using external datasets (e.g., M3 Competition benchmarks) and synthetic data augmentation (p. 14, Sec. 8):

 

"Future iterations will incorporate cross-industry datasets and generative adversarial networks (GANs) to simulate demand volatility, enabling broader validation." (lines 492-493)

 

Clarified in the Discussion (Section 6) that the current work targets SMEs with similar data constraints, serving as a "proof of concept" for scalable adoption (p. 13).

 

 

3. Table 3 title Accuracy Assessment and Biases in Weekly Demand Forecasting is too broad. Something more direct, like 'Comparison of Actual Sales and Forecasted Values' would be clearer. Also, phrases like 'This is a work…' (Line no 49) in the introduction and methodology sections sound unclear and should be rephrased for better readability.

 

Response:

 

We have revised these as follows:

 

Table 3 Title: Changed to "Weekly Sales vs. LSTM Forecast: Deviations (2022–2023)" (p. 11).

 

Line 49 (Introduction): The phrase "This is a work, which lays the groundwork…" was rephrased to:

 

"This study establishes a foundation for future operational implementation, particularly for SMEs seeking scalable, data-driven forecasting."

 

Methodology (Section 4.3): Removed redundant phrasing (e.g., "The Algorithm 1 is the implementation…") and streamlined descriptions for clarity.

 

Typos: Corrected minor errors (e.g., "overslocking" → "overstocking" in Sec. 7).

 

We believe these revisions resolve all outstanding issues. Thank you again for your valuable input, which has strengthened our manuscript. Should further clarifications be needed, we are happy to provide them.

 

Sincerely,

The Corresponding Authors

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for your efforts in revising the manuscript. The changes made have improved certain aspects of clarity and presentation. However, several points still require attention to further strengthen the quality, completeness, and applicability of your work.

1. Discussion vs. Conclusion

• The detailed comparison with ARIMA should be presented in the Discussion section rather than in the Conclusion. In the conclusion, it can be briefly summarized to highlight key takeaways.

2. Figure Presentation

• Figure 3 could be made clearer by adjusting the Y-axis bounds (e.g., from 800 to 1400) to improve visual readability.

3. Missing 2021 Data in Table 3

• While 2021 data are included in Figure 3, they are not reported in Table 3. The omission of 2021 data from the Accuracy Assessment and Biases in Weekly Demand Forecasting must be justified.

4. Language and Style

• Some sentences require revision for grammatical accuracy and clarity. For instance, Lines 435–436:
  “Hybrid approaches for volatile products, combining LSTM with anomaly detection (e.g., Isolation Forests) to flag outliers.”
  This should be rephrased for proper sentence structure and readability.

5. Model Versatility vs. Limitations

• The paper claims that the modular architecture is adaptable across diverse manufacturing sectors (e.g., automotive, electronics, consumer goods). However, the current implementation:
• Does not model complex seasonality patterns
• Ignores external influencing factors such as promotions, market shifts, or macroeconomic indicators
  This gap raises concerns about the practicality and applicability of the model in real-world demand planning. Please address this contradiction or explain why these limitations do not undermine the claimed versatility.

6. Case Study Context and Relevance

• As noted in earlier feedback, the manuscript still lacks sufficient contextual details about the case study. Critical information—such as the product type, industrial sector, and production strategy (e.g., make-to-stock, mass production)—is necessary to assess the generalizability of the results.
• For example, in week 41, the forecast deviation exceeds 22%. Is such an error margin acceptable for SMEs? Providing a detailed case description will help establish the practical relevance and validity of your approach.

Addressing these points will significantly enhance the manuscript’s technical depth, clarity, and real-world applicability.

Regards,

Author Response

Dear Reviewer,

 

Thank you for your feedback, which has further strengthened our manuscript. We have addressed all points below, with revisions highlighted in the updated manuscript.

 

1. Discussion vs. Conclusion

Comment: "Detailed ARIMA comparison belongs in Discussion, not Conclusion."

 

Response:

Moved the quantitative ARIMA comparison (Section 6, Discussion), including MAE reduction (42%) and accuracy metrics (93.2%).

 

Revised Section 8 (Conclusions) to summarize key outcomes concisely:

 

"The LSTM model outperformed traditional methods (e.g., ARIMA) in accuracy and operational cost reduction, establishing a foundation for SME adoption."

 

2. Figure 3 is missing axis labels, units, and a clear legend. The image quality is low, which makes it hard to interpret the trends. The title also feels vague and should better describe what the figure shows.

 

Response:

We agree and have revised Figure 3 as follows:

 

Axis labels: Added "Weeks" (X-axis) and "Units Sold" (Y-axis).

 

Image Quality: Replaced with a high-resolution version.

 

3. Missing 2021 Data in Table 3

Comment: "Justify omitting 2021 data from Table 3."

 

Response:

Added justification in Section 5 (Results):

 

*"Table 3 excludes 2021 sales data to prioritize direct comparison between the most recent historical years (2022–2023) and the forecasted year (2024). This aligns with industry practices for near-term forecast validation."*

 

4. Language and Style

Comment: "Rephrase unclear sentences (e.g., Lines 435–436)."

 

Response:

Revised problematic sentences:

 

Original: "Hybrid approaches for volatile products, combining LSTM with anomaly detection (e.g., Isolation Forests) to flag outliers."

 

Revised: "For volatile products, we recommend hybrid approaches (e.g., LSTM with Isolation Forests) to detect anomalies and flag outliers."

 

5. Model Versatility vs. Limitations

Comment: "Address contradictions in claimed versatility vs. ignored factors (seasonality, promotions, etc.)."

 

Response:

Added Subsection 7.4 (Scope Limitations) to clarify:

 

*"While the model’s modular architecture supports cross-sector deployment, its current iteration focuses on foundational temporal patterns. It does not capture complex seasonality, promotions, or macroeconomic factors—common limitations in SME contexts with sparse data. Future work (Section 8) will integrate these variables."*

 

Emphasized in the Discussion that simplicity is a strategic trade-off for SMEs:

 

"The model prioritizes ease of implementation over complexity, enabling SMEs to adopt AI-driven forecasting without external data dependencies."

 

6. Case Study Context

Comment: "Add product/sector details and justify error margins (e.g., Week 41’s 22% deviation)."

 

Response:

Included case study context in Section 5 (Results):

 

*"The model was tested for an automotive SME producing electrical components (make-to-stock strategy). Week 41’s deviation (-22%) coincided with a supply chain disruption (documented in internal logs), reflecting real-world volatility. For SMEs, deviations <25% remain operationally acceptable, as excess stock costs (15–20%) outweigh shortage risks (30–35% revenue loss [2,28])."*

 

Added Table 4: SME error tolerance thresholds based on industry benchmarks.

 

We added supplementary material

 

Future Work: Expanded Section 8 to include:

 

*"Integration of seasonality (SARIMA-LSTM hybrids) and external factors (e.g., commodity APIs) in v2.0."*

 

We deeply appreciate your review, which has significantly enhanced the manuscript’s rigor and applicability.

 

Sincerely,

The Corresponding Authors

Author Response File: Author Response.pdf

Round 3

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Your revisions demonstrate a thorough engagement with the reviewers’ feedback: the literature review has been appropriately expanded, illustrations and tables have been clarified and standardized, the case study results have been reorganized to highlight key findings, and the discussion more clearly connects those findings to the research questions and the broader field.

While the manuscript’s novelty remains modest, these amendments have elevated it to an acceptable scientific standard. Minor copyediting for language consistency and formatting will suffice to finalize the paper.

Regards,