Financial Statement Fraud Detection Through an Integrated Machine Learning and Explainable AI Framework
Bibliographic record
Abstract
Financial statement fraud remains a substantial risk in environments marked by weak regulatory oversight and information asymmetry. This study develops a decision-centric framework that integrates machine learning, explainable artificial intelligence, and decision curve analysis to improve fraud detection under severe class imbalance. Using 969 firm-year observations from 132 Mongolian firms (2013–2024), we evaluate 21 financial ratios with models including Random Forest, XGBoost, LightGBM, MLP, TabNet, and a Stacking Ensemble trained with SMOTE and class-weighted learning. Performance was assessed using PR-AUC, F1-score, Recall, and DeLong-based significance testing. The Stacking Ensemble achieved the strongest results (PR-AUC = 0.93; F1 = 0.83), outperforming both classical and modern baseline models. Interpretability analyses (SHAP, LIME, and counterfactual explanations) consistently identified leverage, profitability, and liquidity indicators as dominant drivers of fraud risk, supported by a SHAP Stability Index of 0.87. Decision curve analysis showed that calibrated thresholds improved decision efficiency by 7–9% and reduced over-audit costs by 3–4%, while an audit cost simulation estimated annual savings of 80–100 million MNT. Overall, the proposed ML–XAI–DCA framework offers a transparent, interpretable, and cost-efficient approach for enhancing fraud detection in emerging-market contexts with limited textual disclosures.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.009 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".