A Machine Learning and Deep Learning-Based Account Code Classification Model for Sustainable Accounting Practices
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsIn may opinion the manuscript is acceptable. Please find below my considerations and a suggestions.
1. The objective of the research is to utilize image processing techniques to predict cash codes in accounting reports, automate accounting processes, and enhance precision. This paper aims to address existing gaps in the literature regarding the use of image-processing techniques in the field of accounting. It also aims to provide guidance to future researchers on how these techniques can be applied in accounting reporting.
2. The paper is well structured and the research hypothesis is clearly formulated.
3. The literature review provides a comprehensive and well-structured summary of the most relevant recent literature on the topic.
4. The methodology and research algorithm are clearly and appropriately delineated, aligning with the objectives and scope of the investigation.
5. The paper is structured in a logical manner. The information and content are effectively organized.
6. The results are pertinent to the study. The research hypotheses are adequately substantiated.
7. The conclusions are suscintly and clearly presented. However, some minor improvements could be made. It would be beneficial to introduce some considerations (based on the results of the research) regarding the mechanisms/solutionfs for the integration of research results into sustainable accounting practices and adaptation to current business models. Additionally, presenting the main externalities to complement the synthesis of benefits in the conclusions would add aditional value to the study.
Author Response
Comments 1: [The conclusions are suscintly and clearly presented. However, some minor improvements could be made. It would be beneficial to introduce some considerations (based on the results of the research) regarding the mechanisms/solutionfs for the integration of research results into sustainable accounting practices and adaptation to current business models. Additionally, presenting the main externalities to complement the synthesis of benefits in the conclusions would add aditional value to the study].
Response 1: [Firstly, we would like to thank you for your valuable feedback and suggestions aimed at improving our research. In the Conclusions section of the study, we have provided detailed evaluations regarding the integration of sustainable accounting practices and current business models. Additionally, the key external impacts of the study's findings on environmental and social sustainability have been thoroughly addressed in the conclusions section.]
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe article explores the application of machine learning and deep learning techniques for automating the classification of accounting account codes using image processing. Various deep learning models, such as Inception V3, VGG-16, and SqueezeNet, are used to extract features from images of financial transactions, which are then classified using algorithms like Neural Networks and Logistic Regression. The study demonstrates that this approach significantly improves the accuracy, efficiency, and sustainability of accounting processes, achieving a classification accuracy of 99.2%.
The introduction of the article highlights the importance of accurately classifying accounting account codes to ensure efficient and error-free financial reporting. It discusses how current accounting processes are often manual, prone to errors, and time-consuming, thus creating the need for automated solutions. The study proposes using image processing techniques and deep learning models to automate the classification of accounting codes, with a focus on improving sustainability in accounting practices. In the first section, the authors delve into the role of image processing in accounting, particularly in reducing errors and enhancing the efficiency of document management. They explain how technologies like Optical Character Recognition (OCR) and Artificial Intelligence (AI) can be used to digitize financial documents and streamline accounting processes. The section also emphasizes the potential of these technologies to support businesses in achieving sustainability goals by reducing manual labor, enhancing accuracy, and enabling paperless workflows.
The second section of the article provides a literature review on the use of image processing techniques in accounting. It discusses how these technologies, particularly Optical Character Recognition (OCR) and AI, are applied to digitize financial documents like invoices and receipts, improving the speed, accuracy, and sustainability of accounting processes. The section also highlights previous studies that have used machine learning and deep learning models to automate tasks such as document recognition and anomaly detection in financial data.
The third section of the article outlines the research methodology, focusing on the use of image processing techniques to classify accounting account codes. Various deep learning models, including Inception V3, VGG-16, and SqueezeNet, were employed to extract features from images of cash and bank transactions. These extracted features were then classified using several machine learning algorithms, such as Neural Networks, Logistic Regression, and k-Nearest Neighbors (kNN). The section also describes the hardware and software used for the experiments, as well as the process of dataset creation and feature extraction from the images.
The fourth section presents the results of the study, comparing the performance of various deep learning models and machine learning algorithms in classifying accounting account codes. Inception V3 combined with a Neural Network achieved the highest classification accuracy of 99.2%, outperforming other models like VGG-16 and SqueezeNet. The section also discusses the effectiveness of different sampling techniques, with random sampling providing the best performance in the classification tasks.
The fifth section concludes that machine learning and deep learning techniques can significantly improve the accuracy and efficiency of accounting processes, particularly in classifying account codes. The study demonstrated that the combination of deep learning models like Inception V3 with Neural Networks is highly effective, achieving a classification accuracy of 99.2%. The authors suggest that future research should explore larger datasets and more advanced models to further enhance the scalability and sustainability of these methods in real-world accounting practices.
The study is limited by the relatively small dataset, which may not fully represent the diversity of financial transactions encountered in real-world accounting scenarios, affecting the generalizability of the results.
The research focuses only on commonly used deep learning and machine learning models, without exploring newer or more advanced techniques that could potentially offer better performance or efficiency.
The study primarily evaluates model accuracy, but does not address the practical integration of these models into real-time accounting systems or their impact on decision-making processes in businesses.
Despite the mentioned limitations, the article presents a valuable contribution to the field by demonstrating how machine learning and deep learning techniques can enhance the accuracy and efficiency of accounting processes. The high classification accuracy achieved, along with the potential for automation and sustainability in accounting, highlights the relevance and applicability of the research. Therefore, the article provides significant insights and can be considered for publication, as it opens avenues for further research and practical advancements in accounting automation.
Author Response
Comments 1: [The study primarily evaluates model accuracy, but does not address the practical integration of these models into real-time accounting systems or their impact on decision-making processes in businesses.]
Response 1: [We would like to thank you for your interest and valuable suggestions. In line with the feedback from our reviewers, and as mentioned in the limitations of our study, we aim to address these limitations in future research. In this context, we plan to conduct future studies using more advanced models such as Transformers and Graph Neural Networks, with larger and more diverse accounting datasets to perform real-time analyses. Additionally, we aim to integrate these models into existing accounting software through API connections, enabling real-time data processing in accounting systems.]
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe study assesses the efficiency/accuracy of using various deep learning and machine learning techniques in accounting processes.
Specifically, both data collection techniques (image processing) and specific learning algorithms (Logistic Regression, Gradient Boosting, Neural 15 Network, kNN, Naive Bayes, and Stochastic Gradient Descent) are considered.
The authors conclude that the use of AI and deep learning techniques in accounting processes improves the accuracy of labelling, increases the speed of document processing, reduces the workload of accountants, facilitates digital archiving and increases operational efficiency. All these facilities contribute to strengthening sustainability in accounting.
The analyzes are limited to a specific set of operations (cash inflows and outflows), but the methodology can be replicated for analyzes at the level of different accounting operations. This situation was considered by the authors as an opportunity to develop future research.
The topic addressed is current, and the results of the study are useful for both researchers and companies.
From the point of view of literature review and empirical research, the study has the potential to contribute to the advancement of knowledge.
Overall, this article is clearly written. Some minor revisions are needed regarding the research methodology.
For more details, see the comments below.
1. The abstract responds to editorial requirements.
2. The introduction places the debate in a wider context, briefly reviews the current state of the research field and cites key scientific publications.
3. The literature review captures current research directions and highlights the research gap regarding the use of image processing techniques in the field of accounting. The authors must review the numbering of subsections (e.g. Subsection 3.1.1 from Section 2.1).
4. The research methodology is properly detailed, facilitating replication. In order to provide more clarity and robustness to the analyses, authors must consider the consistent use of specific terminology (e.g.: regression analysis vs. “regresyon lojistik” - Figure 2).
5. The results of the research are discussed with accuracy. To increase the value of the paper, the authors may consider scoring the extent to which the results of their study (“the highest model performance was achieved with the Neural Network model using the Inception V3 feature extractor and 10 random samplings 453 with a 70% dataset”) are convergent with the results of previous research. The results of the studies already cited (e.g. [51-57]) which analyzed the efficiency of the digital transfer of accounting documents (e.g.: receipts, invoices, and delivery notes) may be taken into account.
6. The conclusions are consistent with the evidence and arguments presented in the previous sections.
Author Response
Comments 3: [3. The literature review captures current research directions and highlights the research gap regarding the use of image processing techniques in the field of accounting. The authors must review the numbering of subsections (e.g. Subsection 3.1.1 from Section 2.1).]
Response 3: [Thank you very much for your interest in our work, as well as for your valuable contributions and suggestions. The section numbering has been corrected, and these changes can be found on page 7, line 129 of the manuscript.]
Comments 4: [4. The research methodology is properly detailed, facilitating replication. In order to provide more clarity and robustness to the analyses, authors must consider the consistent use of specific terminology (e.g.: regression analysis vs. “regresyon lojistik” - Figure 2).]
Response 4: [The typographical error in Figure 2 has been corrected. Additionally, the entire manuscript has been meticulously reviewed for similar errors. These corrections can be found on page 9, line 201 of the manuscript.]
Comments 5: [The results of the research are discussed with accuracy. To increase the value of the paper, the authors may consider scoring the extent to which the results of their study (“the highest model performance was achieved with the Neural Network model using the Inception V3 feature extractor and 10 random samplings 453 with a 70% dataset”) are convergent with the results of previous research. The results of the studies already cited (e.g. [51-57]) which analyzed the efficiency of the digital transfer of accounting documents (e.g.: receipts, invoices, and delivery notes) may be taken into account.]
Response 5: [Thank you very much for your suggestion. In line with your recommendation, necessary additions have been made in the Conclusions section of the study, evaluating the alignment of our results with previous research.]
Author Response File: Author Response.pdf
