Predicting financial reports fraud by machine learning: the proxy of auditor opinions

Abstract

This study examines the effectiveness of various machine learning algorithms in detecting financial report fraud among firms listed on the Ho Chi Minh City and Hanoi stock exchanges. Using 2,747 observations from 347 firms between 2010 and 2023 (sourced from Refinitiv), the study compares two fraud proxies—M-Score and auditor opinion. Algorithms evaluated include Logistic Regression, KNN, Naive Bayes, SVM (linear and sigmoid), Decision Tree, Random Forest, ANN, and XGBoost. Predictor variables are selected based on the Fraud Triangle and Fraud Diamond theories, incorporating factors such as revenue growth, return on assets, leverage, auditor quality, and executive compensation. Results show that Random Forest and XGBoost achieve the highest accuracy across both fraud proxies. The findings demonstrate the potential of machine learning to enhance fraud detection, especially in emerging markets. This study is among the first to empirically compare multiple algorithms using dual fraud proxies and to integrate the full Fraud Diamond framework in variable selection. The results offer valuable insights for investors, corporate managers, and regulators aiming to strengthen financial oversight and reduce fraud risk.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 15 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.