Minh Nguyen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Minh Nguyen has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
Bang Nguyen
Faculty of Data Science University of Finance ‐ Marketing Ho Chi Minh City Vietnam
2 shared publications- Quang-Thinh Bui
Applied Mathematics and Data Science (AMaDaS) Research Group, Tien Giang University, Dong Thap, Vietnam
2 shared publications - Anh Minh Nguyen1 shared publication
- Duong Q. Nguyen
Center for AI Research, Vin University, Hanoi, Vietnam
1 shared publication - Farzana Nasrin
University of Hawaii at Manoa
1 shared publication - Hiep Huynh
Tien Giang University
1 shared publication
Latest Publications
- Open Access
- 20 October 2026
- Published by Elsevier in SSRN Electronic Journal
Cheap Production, Costly Verification: Doing Research under Agentic AI and Automation
Artificial intelligence (AI) is making research cheap to produce but not (yet) cheap to check. AI agents can now draft manuscripts, run analyses, and even generate mathematical proofs at a small fraction of their former cost, whereas verifying a claim, by checking a proof, replicating an experiment, or auditing data, still relies largely on scarce expert time. We develop a mathematical model of the research system in which researchers decide whether to produce and submit claims, and a verifier with limited capacity decides which claims to check. As production becomes cheaper, more claims are submitted, their average reliability falls because cost no longer filters out low-effort work, and a growing share of claims goes unchecked. We show that the flow of verified discoveries first rises and then falls as automation increases, a pattern we call the Verification Laffer Curve, unless verification capacity grows in step with submission volume, and that review backlogs and the time to validation can grow without bound once submissions outpace reviewing capacity. The model further predicts that, even with no difference in ability and no reviewer bias, the work of early-career researchers and of researchers working outside their home field receives less verification because it is costlier to check, and that this gap widens as automation grows. Numerical simulations illustrate these results. We discuss how expanding verification capacity, restoring screening, and directly lowering verification costs for disadvantaged researchers can keep AI a net gain for science.
- 20 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessCheap Production, Costly Verification: Doing Research under Agentic AI and Automation
Artificial intelligence (AI) is making research cheap to produce but not (yet) cheap to check. AI agents can now draft manuscripts, run analyses, and even generate mathematical proofs at a small fraction of their former cost, whereas verifying a claim, by checking a proof, replicating an experiment, or auditing data, still relies largely on scarce expert time. We develop a mathematical model of the research system in which researchers decide whether to produce and submit claims, and a verifier with limited capacity decides which claims to check. As production becomes cheaper, more claims are submitted, their average reliability falls because cost no longer filters out low-effort work, and a growing share of claims goes unchecked. We show that the flow of verified discoveries first rises and then falls as automation increases, a pattern we call the Verification Laffer Curve, unless verification capacity grows in step with submission volume, and that review backlogs and the time to validation can grow without bound once submissions outpace reviewing capacity. The model further predicts that, even with no difference in ability and no reviewer bias, the work of early-career researchers and of researchers working outside their home field receives less verification because it is costlier to check, and that this gap widens as automation grows. Numerical simulations illustrate these results. We discuss how expanding verification capacity, restoring screening, and directly lowering verification costs for disadvantaged researchers can keep AI a net gain for science.
- 17 August 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
p -Hacking and Publication Bias in Design-Based Causal Studies in Information Systems
Design-based causal methods such as difference in differences, instrumental variables, randomized controlled trials, regression discontinuity, and synthetic control have become the foundation of evidence-based decision making in information systems (IS). Their value, however, depends on the credibility of the published evidence. Analyzing 7,516 hypothesis tests from 558 articles published in seven leading IS journals between 2000 and 2023, we identify systematic clustering of statistical results immediately above the conventional 5% significance threshold, particularly for difference in differences, instrumental variables, and randomized controlled trials. These patterns suggest that publication incentives and researcher degrees of freedom may distort the evidence base used by managers and policymakers. We translate these findings into practical recommendations for researchers, reviewers, and editors. Specifically, we advocate greater transparency in research design and reporting, stronger requirements for data and code availability, broader acceptance of well-executed null findings, routine sensitivity analyses, and method-specific reporting standards for causal studies. We also introduce an IS-calibrated local false discovery rate that helps readers assess the credibility of reported causal estimates. By strengthening evidence quality rather than discouraging causal research, these recommendations aim to improve the reliability, reproducibility, and policy relevance of empirical findings that increasingly guide organizational investment, digital transformation, platform governance, and public policy decisions.
- 17 August 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
p -Hacking and Publication Bias in Design-Based Causal Studies in Information Systems
Design-based causal methods such as difference in differences, instrumental variables, randomized controlled trials, regression discontinuity, and synthetic control have become the foundation of evidence-based decision making in information systems (IS). Their value, however, depends on the credibility of the published evidence. Analyzing 7,516 hypothesis tests from 558 articles published in seven leading IS journals between 2000 and 2023, we identify systematic clustering of statistical results immediately above the conventional 5% significance threshold, particularly for difference in differences, instrumental variables, and randomized controlled trials. These patterns suggest that publication incentives and researcher degrees of freedom may distort the evidence base used by managers and policymakers. We translate these findings into practical recommendations for researchers, reviewers, and editors. Specifically, we advocate greater transparency in research design and reporting, stronger requirements for data and code availability, broader acceptance of well-executed null findings, routine sensitivity analyses, and method-specific reporting standards for causal studies. We also introduce an IS-calibrated local false discovery rate that helps readers assess the credibility of reported causal estimates. By strengthening evidence quality rather than discouraging causal research, these recommendations aim to improve the reliability, reproducibility, and policy relevance of empirical findings that increasingly guide organizational investment, digital transformation, platform governance, and public policy decisions.
- Open Access
- 30 March 2026
- Published by Elsevier in SSRN Electronic Journal
Flying to the Cruise-Port in Miami: Predicting Passenger Satisfaction in Cruise-Linked Air Travel Using Machine Learning
This study investigates the operational drivers of airline passenger satisfaction within the specific setting of cruise-linked air travel to the Port of Miami. Using an anonymized survey dataset that combines service-quality ratings, passenger attributes, and a verified satisfaction label, we benchmark seven supervised classifiers, including Logistic Regression, Linear SVM, Decision Tree, Random Forest, XGBoost, LightGBM, and CatBoost. To test how information is represented and compressed, we evaluate three feature designs: (i) fourteen service-rating factors, (ii) a single engineered Total Satisfaction score formed as the sum of the fourteen factors, and (iii) the combined representation (fourteen factors plus the aggregate score). Models are evaluated using accuracy, balanced accuracy, precision, recall, F1-score, ROC-AUC, ROC curves, and confusion matrices. Across models, the representation based on the fourteen service factors yields strong discrimination, with boosting/ensemble methods achieving near-ceiling ROC-AUC (approximately 0.99) and accuracy around 0.95. In contrast, compressing the experience into a single Total Satisfaction feature produces substantially weaker discrimination (ROC-AUC $\approx 0.79$) and causes most models to converge to nearly identical ROC behavior, indicating that representational compression, rather than model choice, becomes the bottleneck. Finally, using Shapley Additive Explanations (SHAP), we show that top three features importance are Inflight wifi service, Online boarding, and Departure/Arrival time convenient. This top-three feature importance is consistent under passengers' age, gender, travel class, type of travel, and customer type. This implies the robust evidence that inflight connectivity and digital touchpoints are core drivers of satisfaction for cruise-bound travelers.
- 30 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFlying to the Cruise-Port in Miami: Predicting Passenger Satisfaction in Cruise-Linked Air Travel Using Machine Learning
This study investigates the operational drivers of airline passenger satisfaction within the specific setting of cruise-linked air travel to the Port of Miami. Using an anonymized survey dataset that combines service-quality ratings, passenger attributes, and a verified satisfaction label, we benchmark seven supervised classifiers, including Logistic Regression, Linear SVM, Decision Tree, Random Forest, XGBoost, LightGBM, and CatBoost. To test how information is represented and compressed, we evaluate three feature designs: (i) fourteen service-rating factors, (ii) a single engineered Total Satisfaction score formed as the sum of the fourteen factors, and (iii) the combined representation (fourteen factors plus the aggregate score). Models are evaluated using accuracy, balanced accuracy, precision, recall, F1-score, ROC-AUC, ROC curves, and confusion matrices. Across models, the representation based on the fourteen service factors yields strong discrimination, with boosting/ensemble methods achieving near-ceiling ROC-AUC (approximately 0.99) and accuracy around 0.95. In contrast, compressing the experience into a single Total Satisfaction feature produces substantially weaker discrimination (ROC-AUC $\approx 0.79$) and causes most models to converge to nearly identical ROC behavior, indicating that representational compression, rather than model choice, becomes the bottleneck. Finally, using Shapley Additive Explanations (SHAP), we show that top three features importance are Inflight wifi service, Online boarding, and Departure/Arrival time convenient. This top-three feature importance is consistent under passengers' age, gender, travel class, type of travel, and customer type. This implies the robust evidence that inflight connectivity and digital touchpoints are core drivers of satisfaction for cruise-bound travelers.
- 1 January 2026
- Published by Springer Nature
- p. 337-351
Measuring Causal Effects of Online Medical Records on Medical Decision Making Using Causal Machine Learning
- 1 January 2026
- Published by Springer Nature
- p. 337-351
Measuring Causal Effects of Online Medical Records on Medical Decision Making Using Causal Machine Learning
- 4 December 2025
- Published by World Scientific Pub Co Pte Ltd in Materials and Emerging Technologies for Sustainability
- Vol. 01 (04)
A novel approach to metaheuristic-driven threshold optimization in frequent high-order pattern mining
Frequent pattern mining is a fundamental task in data mining, although its stability and effectiveness are generally constrained by the sensitivity of the frequency threshold. Simplicial complexes capture higher-order relationships by representing multi-entity interactions beyond simple pairwise connections. When data are represented as simplicial complexes, the combinatorial explosion in the search space further intensifies computational challenges. This study proposes, FreSCo[ ], an enhanced FreSCo framework that incorporates an automatic threshold optimization mechanism driven by Particle Swarm Optimization (PSO). The key objective is to dynamically balance pattern coverage and processing overhead. The proposed framework features adaptive threshold selection, early pruning strategies, caching-based reuse to expedite convergence, and reduce redundant computations. Experimental evaluations on three benchmark datasets, including ETFs, Zebra, and Enron, demonstrate significant performance improvements with average runtime reductions of 37.19% on ETFs, 15.55% on Zebra, and 15.39% on Enron, while maintaining accuracy and stability across 10 independent trials. These results indicate that the PSO-optimized FreSCo framework achieves faster convergence, improved scalability, and increased efficiency in frequent simplicial complex mining. Consequently, this framework establishes a robust foundation for metaheuristic-driven optimization approaches in higher-order data mining.
- 4 December 2025
- Published by World Scientific Pub Co Pte Ltd in Materials and Emerging Technologies for Sustainability
- Vol. 01 (04)
A novel approach to metaheuristic-driven threshold optimization in frequent high-order pattern mining
Frequent pattern mining is a fundamental task in data mining, although its stability and effectiveness are generally constrained by the sensitivity of the frequency threshold. Simplicial complexes capture higher-order relationships by representing multi-entity interactions beyond simple pairwise connections. When data are represented as simplicial complexes, the combinatorial explosion in the search space further intensifies computational challenges. This study proposes, FreSCo[ ], an enhanced FreSCo framework that incorporates an automatic threshold optimization mechanism driven by Particle Swarm Optimization (PSO). The key objective is to dynamically balance pattern coverage and processing overhead. The proposed framework features adaptive threshold selection, early pruning strategies, caching-based reuse to expedite convergence, and reduce redundant computations. Experimental evaluations on three benchmark datasets, including ETFs, Zebra, and Enron, demonstrate significant performance improvements with average runtime reductions of 37.19% on ETFs, 15.55% on Zebra, and 15.39% on Enron, while maintaining accuracy and stability across 10 independent trials. These results indicate that the PSO-optimized FreSCo framework achieves faster convergence, improved scalability, and increased efficiency in frequent simplicial complex mining. Consequently, this framework establishes a robust foundation for metaheuristic-driven optimization approaches in higher-order data mining.
- 8 November 2025
- Published by Springer Nature
- p. 339-354
Topological Framework for Exploring Imbalanced Time Series
- 8 November 2025
- Published by Springer Nature
- p. 339-354
Topological Framework for Exploring Imbalanced Time Series
- 5 August 2024
- Published by Wiley in Journal of Forecasting
- Vol. 43 (8) , 3128-3160
Corporate financial distress prediction in a transition economy
Forecasting financial distress of corporations is a difficult task in economies undergoing transition, as data are scarce and are highly imbalanced. This research tackles these difficulties by gathering reliable financial distress data in the context of a transition economy and employing the synthetic minority oversampling technique (SMOTE). The study employs seven different models, including linear discriminant analysis (LDA), logistic regression (LR), support vector machines (SVMs), neural networks (NNs), decision trees (DTs), random forests (RFs), and the Merton model, to predict financial distress among publicly traded companies in Vietnam between 2011 and 2021. The first six models use accounting‐based variables, while the Merton model utilizes market‐based variables. The findings indicate that while all models perform fairly well in predicting results for nondelisted firms, they perform somewhat poorly in predicting results for delisted firms in terms of various measures including balanced accuracy, Matthews correlation coefficient (MCC), precision, recall, and score. The study shows that the models that incorporate both Altman's and Ohlson's variables consistently outperform those that only use Altman's or Ohlson's variables in terms of balanced accuracy. Additionally, the study finds that NNs are generally the most effective models in terms of both balanced accuracy and MCC. The most important variable in Altman's variables as well as the combination of Altman's and Ohlson's variables is “reat” (retained earnings on total assets), whereas “ltat” (total liabilities on total assets) and “wcapat” (working capital on total assets) are the most important variables in Ohlson's variables. The study also reveals that in most cases, the models perform better in predicting results for big firms than for small firms and typically better than in good years than for bad years.
- 5 August 2024
- Published by Wiley in Journal of Forecasting
- Vol. 43 (8) , 3128-3160
Corporate financial distress prediction in a transition economy
Forecasting financial distress of corporations is a difficult task in economies undergoing transition, as data are scarce and are highly imbalanced. This research tackles these difficulties by gathering reliable financial distress data in the context of a transition economy and employing the synthetic minority oversampling technique (SMOTE). The study employs seven different models, including linear discriminant analysis (LDA), logistic regression (LR), support vector machines (SVMs), neural networks (NNs), decision trees (DTs), random forests (RFs), and the Merton model, to predict financial distress among publicly traded companies in Vietnam between 2011 and 2021. The first six models use accounting‐based variables, while the Merton model utilizes market‐based variables. The findings indicate that while all models perform fairly well in predicting results for nondelisted firms, they perform somewhat poorly in predicting results for delisted firms in terms of various measures including balanced accuracy, Matthews correlation coefficient (MCC), precision, recall, and score. The study shows that the models that incorporate both Altman's and Ohlson's variables consistently outperform those that only use Altman's or Ohlson's variables in terms of balanced accuracy. Additionally, the study finds that NNs are generally the most effective models in terms of both balanced accuracy and MCC. The most important variable in Altman's variables as well as the combination of Altman's and Ohlson's variables is “reat” (retained earnings on total assets), whereas “ltat” (total liabilities on total assets) and “wcapat” (working capital on total assets) are the most important variables in Ohlson's variables. The study also reveals that in most cases, the models perform better in predicting results for big firms than for small firms and typically better than in good years than for bad years.
- Open Access
- 1 January 2023
- Published by Elsevier in SSRN Electronic Journal
Natural Disaster Risk and Firm Performance: Text Mining and Machine Learning Approach
We develop a perceived measure of firms’ disaster exposure and/or preparedness equal to the number of words related to natural disaster events in the firms’ Form 10-Ks. We then link this measure to contemporary and future firm decision-making and performance. We find that this perceived natural disaster risk and the government-reported damages of natural hazards this year are negatively associated with firm profitability next year. However, the perceived natural disaster risk is not associated with sales growth and Tobin's Q ratio. Specifically, the perceived natural disaster risk negatively affects firm profitability in the services sector but not in the manufacturing sector. The firm profitability in the services sector is also negatively affected by the billion-dollar natural disasters in the same year. Finally, we find that advanced machine learning models robustly outperform linear regression in predicting firm performance under natural disaster risks. The main implication from this study is that we can employ textual data in financial reports to measure the perceived natural disaster risk and predict its effects on firm performance.
- 1 January 2023
- Published by Elsevier in SSRN Electronic Journal
Open AccessNatural Disaster Risk and Firm Performance: Text Mining and Machine Learning Approach
We develop a perceived measure of firms’ disaster exposure and/or preparedness equal to the number of words related to natural disaster events in the firms’ Form 10-Ks. We then link this measure to contemporary and future firm decision-making and performance. We find that this perceived natural disaster risk and the government-reported damages of natural hazards this year are negatively associated with firm profitability next year. However, the perceived natural disaster risk is not associated with sales growth and Tobin's Q ratio. Specifically, the perceived natural disaster risk negatively affects firm profitability in the services sector but not in the manufacturing sector. The firm profitability in the services sector is also negatively affected by the billion-dollar natural disasters in the same year. Finally, we find that advanced machine learning models robustly outperform linear regression in predicting firm performance under natural disaster risks. The main implication from this study is that we can employ textual data in financial reports to measure the perceived natural disaster risk and predict its effects on firm performance.
- Open Access
- 1 January 2023
- Published by Elsevier in SSRN Electronic Journal
Corporate Financial Distress Prediction in a Transition Economy
Forecasting financial distress of corporations is a difficult task in economies undergoing transition, as data is scarce and is highly imbalanced. This research tackles these difficulties by gathering reliable financial distress data in the context of a transition economy and employing the synthetic minority oversampling technique (SMOTE). The study employs five different models, including linear discriminant analysis (LDA), logistic regression, support vector machines (SVM), neural networks (NN), and the Merton model, to predict financial distress among publicly-traded companies in Vietnam between 2011 and 2021. The first four models use accounting-based variables, while the Merton model utilizes market-based variables. The findings indicate that while all models perform fairly well in predicting results for non-delisted firms, they perform somewhat poorly in predicting results for delisted firms in terms of various measures such as balanced accuracy, precision, recall, and F1 score. The study shows that the models that incorporate both Altman's and Ohlson's variables consistently outperform those that only use Altman's or Ohlson's variables when it comes to balanced accuracy. Additionally, the study finds that neural networks are consistently the most effective models in terms of both balanced accuracy and Matthews correlation coefficient (MCC). The most important variable in Altman's variables as well as the combination of Altman's and Ohlson's variables is "reat" (retained earnings over total assets), whereas "ltat" (total liabilities over total assets) and "wcapat" (working capital over total assets) are the most important variables in Ohlson's variables. The study also reveals that in most cases, the models perform better in predicting results for big firms than for small firms, and the predictive performance in good years is typically better than in bad years in terms of MCC.
- 1 January 2023
- Published by Elsevier in SSRN Electronic Journal
Open AccessCorporate Financial Distress Prediction in a Transition Economy
Forecasting financial distress of corporations is a difficult task in economies undergoing transition, as data is scarce and is highly imbalanced. This research tackles these difficulties by gathering reliable financial distress data in the context of a transition economy and employing the synthetic minority oversampling technique (SMOTE). The study employs five different models, including linear discriminant analysis (LDA), logistic regression, support vector machines (SVM), neural networks (NN), and the Merton model, to predict financial distress among publicly-traded companies in Vietnam between 2011 and 2021. The first four models use accounting-based variables, while the Merton model utilizes market-based variables. The findings indicate that while all models perform fairly well in predicting results for non-delisted firms, they perform somewhat poorly in predicting results for delisted firms in terms of various measures such as balanced accuracy, precision, recall, and F1 score. The study shows that the models that incorporate both Altman's and Ohlson's variables consistently outperform those that only use Altman's or Ohlson's variables when it comes to balanced accuracy. Additionally, the study finds that neural networks are consistently the most effective models in terms of both balanced accuracy and Matthews correlation coefficient (MCC). The most important variable in Altman's variables as well as the combination of Altman's and Ohlson's variables is "reat" (retained earnings over total assets), whereas "ltat" (total liabilities over total assets) and "wcapat" (working capital over total assets) are the most important variables in Ohlson's variables. The study also reveals that in most cases, the models perform better in predicting results for big firms than for small firms, and the predictive performance in good years is typically better than in bad years in terms of MCC.