Predicting audit qualifications using dynamic selection models and explainable AI: evidence from Indian listed firms

The present study examines the effectiveness of Dynamic Selection (DS) techniques in predicting audit qualifications of Indian listed companies. Traditional machine-learning classifiers use a single global decision rule for all observations, which may not adequately capture the diverse financial conditions associated with audit qualification decisions. To address this limitation, the study compares the performance of Dynamic Classifier Selection (DCS) and Dynamic Ensemble Selection (DES) methods with traditional machine-learning classifiers and static ensemble methods. Furthermore, SHapley Additive exPlanations (SHAP) are employed to improve model transparency and identify the key financial factors influencing audit qualification. The analysis is based on a balanced sample of 730 firm-year observations comprising 365 qualified and 365 unqualified audit opinions from BSE-listed non-financial firms during 2013–2022. The best-performing model was the Randomized Reference Classifier (RRC), a DES model, achieving an Area Under the Curve (AUC) of 0.8389. SHAP analysis identified Total Assets Turnover, Total Debt to Total Assets, Short-Term Debt to Total Liabilities, Diluted EPS, and Net Operating Cash Flow to Total Debt as the most influential predictors. The proposed framework combines adaptive model selection with explainable AI, offering valuable insights for auditors, regulators, and financial statement users.