Neural Networks in Accounting: Bridging Financial Forecasting and Decision Support Systems
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors 1. The author should clarify the difference between this work and his previous work entitled” Unlocking Business Value: Integrating AI-Driven Decision-Making in Financial Reporting Systems”. 2- However, the manuscript currently lacks an actual application that uses real-world data to validate the effectiveness of neural networks in financial and accounting contexts. While the theoretical discussion and methodology are well-presented on simulated data, the absence of empirical validation weakens the practical applicability of the proposed approach to real data. To strengthen the manuscript, the author should consider acquiring real-world datasets from relevant domains such as: E-commerce: customer purchasing behavior, or Travel & Tourism: Demand forecasting, dynamic pricing. once a suitable dataset is obtained, the author should preprocess the data, train the proposed neural network model, and evaluate its performance. a discussion of the results should be included, highlighting key insights, potential limitations, and the implications for financial and accounting professionals. This empirical validation will significantly improve the manuscript’s impact and convince readers of the model’s effectiveness. 3. There is more detail through many sections of the systems such as from line 143 to line 163 that should be summarized so that it is not boring for the reader.
Author Response
Comments 1. The author should clarify the difference between this work and his previous work entitled” Unlocking Business Value: Integrating AI-Driven Decision-Making in Financial Reporting Systems”.
Response 1
We thank the reviewer for addressing the issue of comparing the previous work to this research. The newer article can be seen as a logical progression from the earlier one focusing on a complementary facet of AI’s role in accounting and financial reporting. In this research we tried to narrow in on the technical and methodological aspects of applying neural networks to accounting tasks, particularly in predicting profitability and enhancing decision support, moving from a holistic examination of AI’s transformative potential in financial reporting to a more focused, technical treatment of how advanced machine-learning models can be designed, validated, and put into practice with the help of a simulated reality tool like Tensor Flow.
Comments 2However, the manuscript currently lacks an actual application that uses real-world data to validate the effectiveness of neural networks in financial and accounting contexts. While the theoretical discussion and methodology are well-presented on simulated data, the absence of empirical validation weakens the practical applicability of the proposed approach to real data. To strengthen the manuscript, the author should consider acquiring real-world datasets from relevant domains such as: E-commerce: customer purchasing behavior, or Travel & Tourism: Demand forecasting, dynamic pricing. Once a suitable dataset is obtained, the author should preprocess the data, train the proposed neural network model, and evaluate its performance. a discussion of the results should be included, highlighting key insights, potential limitations, and the implications for financial and accounting professionals. This empirical validation will significantly improve the manuscript’s impact and convince readers of the model’s effectiveness.
Response 2
We thank the reviewer for this great and valuable suggestion and example for data sets. Following your recommendation, we have added a focused empirical component to our methodology. Using a Kaggle dataset on S&P 500 firms, we trained a small neural network to predict share prices based on fundamental variables (e.g., Price/Earnings, Dividend Yield). This part, although modest, demonstrates that our model can indeed handle real‐world complexities beyond synthetic or simulated data. We have updated the Methodology (lines 202-214) and the Results and discussion section of the paper (lines 557-623) to detail these procedures and results obtained using MATLAB. Thank you, the suggestion provides complexity and empirical validation to the research
Comments 3 There is more detail through many sections of the systems such as from line 143 to line 163 that should be summarized so that it is not boring for the reader.
Response 3
Thank you for the suggestion, we summarized and reduced the descriptions from Materials and methods section from lines 153 to 214. We also extended the conclusion section considering new findings from line 640-652 and 659-665
Reviewer 2 Report
Comments and Suggestions for AuthorsThis paper introduces a neural network model for financial forecasting in accounting. It integrates traditional accounting principles with a two-hidden-layer network to capture non-linear relationships among key financial metrics. The study details the model’s architecture with corresponding mathematical formulations and uses synthetic data in TensorFlow Playground to demonstrate issues like overfitting and the positive impact of regularization.
Comments:
1. The motivation of this paper is somewhat diluted by an overreliance on historical narratives (e.g., references to Pacioli) that, while interesting, do not directly justify the need for a new neural network model. The authors are suggested to give more theoretical discussions and an intuitionistic example to strongly motivate the work.
2. The literature review covers a range of topics but lacks a focused, critical discussion of recent state-of-the-art approaches for financial forecasting. The paper could benefit from including references to research on financial forecasting, such as the most recent method, RHINE https://doi.org/10.1137/1.9781611978032.61, which would strengthen the argument and provide readers with additional resources for further exploration.
3. The network design (two hidden layers with specific node counts) appears to be chosen arbitrarily without rigorous theoretical or empirical justification. Also, although the paper provides standard neural network formulas (e.g., weight matrices, bias vectors, activation functions), these are largely conventional and do not contribute novel insights into model design.
4. There is a lack of comparison with competitive baseline methods in the experiments part.
Author Response
Comments 1. The motivation of this paper is somewhat diluted by an overreliance on historical narratives (e.g., references to Pacioli) that, while interesting, do not directly justify the need for a new neural network model. The authors are suggested to give more theoretical discussions and an intuitionistic example to strongly motivate the work.
Response 1. We appreciate and thank you for your feedback regarding our references to historical figures like Luca Pacioli, and we acknowledge that while these anecdotes can be interesting, they may not by themselves clarify the pressing need for a modern neural network in accounting. We modified and extended the theoretical discussions with additional examples to motivate the research from Lines 60-79
Comments 2
The literature review covers a range of topics but lacks a focused, critical discussion of recent state-of-the-art approaches for financial forecasting. The paper could benefit from including references to research on financial forecasting, such as the most recent method, RHINE https://doi.org/10.1137/1.9781611978032.61, which would strengthen the argument and provide readers with additional resources for further exploration.
Response 2. Thank you for this suggestion. We included the reference in the literature review, (lines 119-124) and I want to thank you again for the suggestion that helped situate our research among the latest developments in the literature when addressing multi-series regime switching in financial forecasting
Comments 3 The network design (two hidden layers with specific node counts) appears to be chosen arbitrarily without rigorous theoretical or empirical justification. Also, although the paper provides standard neural network formulas (e.g., weight matrices, bias vectors, activation functions), these are largely conventional and do not contribute novel insights into model design.
Response 3. Thank you for your valuable insight and suggestion. Indeed, we neglected to highlight why we chose to use two hidden layers. We addressed the issue in lines 319-339 but the empirical justification exists. We appreciate your feedback regarding our choice of two hidden layers and the standard neural network formulas. Our main objective was to tailor a modestly deep yet interpretable architecture to the complexity inherent in cross-framework financial data. In particular, corporate reporting under IFRS and US GAAP can vary in the way certain revenues and expenses are classified, creating subtle but important differences in the underlying numerical relationships. By employing two hidden layers, rather than a single, more limited layer or an excessively deep network, we find a middle ground that can capture non-linear interactions among distinct financial metrics without unduly complicating training or interpretability
Reviewer 3 Report
Comments and Suggestions for AuthorsThe research is logical and consistent. The authors set a thesis that they prove. The experiments performed and the change in the network model during the research are very well described. But it seems to me that it is good to conduct a study with real data and thus show the practical benefits of the research, and not just a confirmation of the theoretical hypothesis of the authors.
Comments for author File: Comments.pdf
Author Response
Comments. The research is logical and consistent. The authors set a thesis that they prove. The experiments performed and the change in the network model during the research are very well described. But it seems to me that it is good to conduct a study with real data and thus show the practical benefits of the research, and not just a confirmation of the theoretical hypothesis of the authors.
Response. We thank the reviewer for the appreciation related to the structure and novelty of the research. Thank you for the suggestion to conduct a study with real data that reveals the practical benefits. It is a suggestion that brings added value to our research. We have now supplemented this with an empirical component using a Kaggle dataset on S&P 500 companies. In MATLAB, we loaded the data, performed an 80/20 train–test split, trained a modest feedforward network, and evaluated performance with mean squared error and root mean squared error. (lines 557-623) This practical validation -addresses the reviewer’s thoughtful suggestion to include real-world evidence and shows that our neural network framework can handle authentic financial figures.
Author Response File: Author Response.docx
Reviewer 4 Report
Comments and Suggestions for AuthorsThis manuscript presents an insightful study on the integration of neural networks in accounting for financial forecasting and decision-making. The research is relevant and aligns with current trends in artificial intelligence applications within finance.
Strengths:
Well-structured introduction providing historical and theoretical context.
Appropriate research design with a clear methodological approach.
Adequate description of methods, including model architecture and training process.
Novel application of neural networks to financial forecasting.
Reasonably well-supported conclusions based on results.
Areas for Improvement:
Clarify research design choices, particularly the use of synthetic data and how it compares to real-world applicability.
Improve readability by simplifying some complex and lengthy sentences.
Enhance discussion by linking findings more directly to hypotheses and prior studies.
Strengthen the practical implications by discussing real-world applications for financial professionals.
Refine references and citations, ensuring all key recent works are included.
Add a limitations section to acknowledge constraints and suggest areas for future research.
Comments for author File: Comments.pdf
Author Response
Comments 1. Clarify research design choices, particularly the use of synthetic data and how it compares to real-world applicability.
Response 1. Thank you for your valuable feedback and suggestions. We added some context in order to clarify our design choices (lines 119-124 and 319-339) and added empirical data to confer the model real world applicability (lines 557- 623)
Comments 2. Improve readability by simplifying some complex and lengthy sentences.
Response 2. Thank you for your valuable suggestion. We have restructured some complex phrases and redesign some descriptions especially in the introduction and literature review sections (lines 60-79, 100-124)
Comments 3. Enhance discussion by linking findings more directly to hypotheses and prior studies.
Response 3. Thank you for suggesting this. We added some prior studies to support the theoretical hypothesis formulated (lines 60-71 and 118-124)
Comments 4. Strengthen the practical implications by discussing real-world applications for financial professionals.
Response 4. Thank you for this suggestion. You are absolutely right. Our model lacked real world application witch we addressed in the results and discussion and conclusion section (line 557-617 and 659-665)
Comments 5. Refine references and citations, ensuring all key recent works are included.
Response 5. Thank you, we refined and added some additional research that better include recent works
Comments 6. Add a limitations section to acknowledge constraints and suggest areas for future research.
Response 6. This is a great idea; we added constraints and future research in the end of the conclusion section lines 640-664
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI thank the author for considering my all comments
Reviewer 2 Report
Comments and Suggestions for AuthorsThanks for the responses. I have carefully read the revision, and I think this paper can be accepted.