Detecting Accounting Fraud in China A‐Share Market With <scp>PU</scp> Learning

ABSTRACT Detecting accounting fraud is critical for financial market integrity. Traditional methods struggle due to imbalanced data and incomplete labelling. This study introduces positive and unlabelled (PU) learning to enhance detection accuracy using labelled fraud cases and extensive unlabelled samples. Analysing China's A‐share market data from 2001 to 2022, segmented into stability (2001–2019) and transition (2020–2022) regulatory periods, results confirm that PU learning significantly improves the detection of fraudulent firms, demonstrating robust performance under varying regulatory and economic conditions. The findings highlight PU learning's effectiveness as a robust method for detecting fraud, offering practical insights for enhancing regulatory oversight and maintaining market stability.