EXPLAINABLE MACHINE LEARNING FOR TAX REPORTING ANOMALY DETECTION AND COMPLIANCE RISK ASSESSMENT

Authors

  • Qiao Xu (Corresponding Author) The University of Delaware, Newark 19716, DE, USA.

Keywords:

Tax compliance, Audit selection, Anomaly detection, Explainable AI, Shapley values, Conformal prediction, False discovery rate

Abstract

Tax administrations must rank large populations of filings for audit with scarce, delayed and selectively observed labels, and every selection has to be justified to the taxpayer. This paper presents TRACE, an explainable framework for tax reporting anomaly detection and compliance risk assessment. TRACE combines domain-informed tax risk features (tax burden, inter-period consistency and digit conformity), peer-relative normalisation within each fiscal-year cross-section and a dual-channel detector that fuses balanced-bagging gradient boosting with an ECOD-style tail-evidence channel. The fused risk score decomposes exactly into feature attributions and six auditable reason codes, and conformal p-values turn it into audit selections with a controlled false-alarm rate or false discovery rate. In an out-of-time evaluation on 146,045 real U.S. firm-years with SEC enforcement labels (test years 2003–2008), TRACE attains the highest AUPRC among ten detectors and an AUC of 0.704, within 0.007 of the RUSBoost benchmark and 0.034 and 0.065 above XGBoost and LightGBM. On a semi-synthetic benchmark that injects six tax misreporting schemes into real financial statements, TRACE gives the best AUC with known (0.861) and unseen (0.761) schemes, and its top reason code identifies the manipulated risk group in 72.5% of known-scheme cases, compared with 39.0% for LIME applied to the same score. On real public-sector audit data, its exact attributions agree more closely with the Shapley values of the auditor's own risk formula than KernelSHAP and LIME. Conformal calibration keeps the false-alarm rate on compliant filers within 0.006 of the nominal level across six test years, and Benjamini–Hochberg selection attains a realised FDR of 0.039 at α = 0.05 with seven times the power of a LightGBM selector.

References

[1] Frank M M, Lynch L J, Rego S O. Tax reporting aggressiveness and its relation to aggressive financial reporting. The Accounting Review, 2009, 84(2): 467-496.

[2] Nigrini M J. A taxpayer compliance application of Benford's law. Journal of the American Taxation Association, 1996, 18(1): 72-91.

[3] Amiram D, Bozanic Z, Rouen E. Financial statement errors: Evidence from the distributional properties of financial statement numbers. Review of Accounting Studies, 2015, 20(4): 1540-1593.

[4] Dechow P M, Ge W, Larson C R, et al. Predicting material accounting misstatements. Contemporary Accounting Research, 2011, 28(1): 17-82.

[5] Bao Y, Ke B, Li B, et al. Detecting accounting fraud in publicly traded U.S. firms using a machine learning approach. Journal of Accounting Research, 2020, 58(1): 199-235.

[6] Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 2019, 1(5): 206-215.

[7] de Roux D, Perez B, Moreno A, et al. Tax fraud detection for under-reporting declarations using an unsupervised machine learning approach. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018: 215-222.

[8] Vanhoeyveld J, Martens D, Peeters B. Value-added tax fraud detection with scalable anomaly detection techniques. Applied Soft Computing, 2020, 86: 105895.

[9] Savić M, Atanasijević J, Jakovetić D, et al. Tax evasion risk management using a hybrid unsupervised outlier detection method. Expert Systems with Applications, 2022, 193: 116409.

[10] Lundberg S M, Lee S I. A unified approach to interpreting model predictions. In: Advances in Neural Information Processing Systems 30 (NIPS), 2017: 4765-4774.

[11] Lundberg S M, Erion G, Chen H, et al. From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2020, 2(1): 56-67.

[12] Bates S, Candès E, Lei L, et al. Testing for outliers with conformal p-values. The Annals of Statistics, 2023, 51(1): 149-178.

[13] Hooda N, Bawa S, Rana P S. Fraudulent firm classification: A case study of an external audit. Applied Artificial Intelligence, 2018, 32(1): 48-64.

[14] Liu F T, Ting K M, Zhou Z H. Isolation forest. In: Proceedings of the 8th IEEE International Conference on Data Mining (ICDM), 2008: 413-422.

[15] Xu H, Pang G, Wang Y, et al. Deep isolation forest for anomaly detection. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(12): 12591-12604.

[16] Li Z, Zhao Y, Botta N, et al. COPOD: Copula-based outlier detection. In: Proceedings of the IEEE International Conference on Data Mining (ICDM), 2020: 1118-1123.

[17] Li Z, Zhao Y, Hu X, et al. ECOD: Unsupervised outlier detection using empirical cumulative distribution functions. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(12): 12181-12193.

[18] Zhao Y, Hryniewicki M K. XGBOD: Improving supervised outlier detection with unsupervised representation learning. In: Proceedings of the International Joint Conference on Neural Networks (IJCNN), 2018: 1-8.

[19] Seiffert C, Khoshgoftaar T M, Van Hulse J, et al. RUSBoost: A hybrid approach to alleviating class imbalance. IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, 2010, 40(1): 185-197.

[20] Chen T, Guestrin C. XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 785-794.

[21] Ke G, Meng Q, Finley T, et al. LightGBM: A highly efficient gradient boosting decision tree. In: Advances in Neural Information Processing Systems 30 (NIPS), 2017: 3146-3154.

[22] Ribeiro M T, Singh S, Guestrin C. 'Why should I trust you?' Explaining the predictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 1135-1144.

[23] Vovk V, Gammerman A, Shafer G. Algorithmic Learning in a Random World. New York, NY, USA: Springer, 2005.

[24] Angelopoulos A N, Bates S. Conformal prediction: A gentle introduction. Foundations and Trends in Machine Learning, 2023, 16(4): 494-591.

[25] Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society: Series B (Methodological), 1995, 57(1): 289-300.

Downloads

Published

2023-09-01

How to Cite

Qiao Xu. Explainable Machine Learning For Tax Reporting Anomaly Detection And Compliance Risk Assessment. World Journal of Economics and Business Research. 2023, 1(1): 56-66. DOI: https://doi.org/10.61784/wjebr4127.