Bagging or boosting? Empirical evidence from financial statement fraud detection

Abstract

AbstractEnsemble learning, specifically bagging and boosting, has been widely used in the financial field for detecting financial fraud, but their relative performance still lacks consensus. This study compares the performance of five ensemble learning models based on bagging and boosting, using data from Chinese A‐share listed companies from 2012 to 2022, including the COVID‐19 pandemic period. Results show that bagging outperforms boosting in various evaluation indicators, with profitability and asset quality positively affecting financial fraud. This study reveals the mechanism by which ensemble learning affects financial fraud detection and expands related research in the financial field.