Sumit Shekhar Reviewer
Approved
Relevance and Originality
The manuscript addresses a highly pertinent problem in contemporary supply chain analytics by combining federated learning with strong privacy protection for collaborative demand forecasting. The focus on automotive supply chains is appropriate and well justified, given the competitive sensitivity and regulatory constraints associated with this sector. The integration of hybrid encryption, adaptive differential privacy, and incentive mechanisms provides a coherent conceptual contribution that goes beyond a purely technical treatment. While individual components of the framework have been studied previously, their coordinated application to privacy preserving collaboration in automotive forecasting represents a meaningful and timely extension of existing research.
Methodology
The methodological framework is clearly articulated and covers system architecture, encryption design, privacy mechanisms, incentive structures, and communication efficiency. The layered description of hybrid encryption and secure aggregation demonstrates careful attention to both security and scalability. The experimental setup is reasonably described, including baseline comparisons and evaluation metrics related to forecasting accuracy and operational impact. However, the study would benefit from more explicit detail on data preprocessing steps, parameter selection for privacy budgets, and the criteria used to configure the adaptive noise mechanism. Additional clarification on how participant heterogeneity was handled during training would also improve reproducibility.
Validity and Reliability
The manuscript provides a thoughtful discussion of potential threats to validity, particularly with respect to privacy leakage, collusion resistance, and cross jurisdictional compliance. The reported use of attack simulations and ablation analysis supports the credibility of the findings. Nevertheless, much of the evidence is drawn from a single industrial dataset, which may limit generalizability to other supply chain contexts. The reliability of conclusions regarding regulatory compliance would be strengthened by clearer linkage between technical guarantees and specific regulatory requirements. A more explicit acknowledgment of possible bias introduced by data imbalance across participants would further enhance this section.
Clarity and Structure
The paper is logically organized, progressing from conceptual motivation to technical design, results, and future directions. Sectioning is clear and the narrative maintains coherence across technical and applied discussions. Figures and tables help summarize complex relationships among privacy, performance, and communication efficiency. Some sections, particularly within the methods and discussion, are dense and could benefit from tighter phrasing and shorter paragraphs. Minor language refinement would improve readability without altering the technical content.
Results and Analysis
The results indicate improved forecasting accuracy and reduced inventory related inefficiencies compared with both isolated and conventional collaborative approaches. The analysis appropriately links technical mechanisms such as adaptive privacy calibration and gradient quantization to observed performance gains. The inclusion of computational and communication efficiency metrics strengthens the practical relevance of the study. However, several performance claims are presented in strongly positive terms and would benefit from more cautious interpretation, especially where projections about large scale adoption are made. A more explicit comparison with non encrypted federated learning under identical conditions would further clarify the tradeoffs between privacy and utility.
Sumit Shekhar Reviewer