Audit opinions are critical elements of financial reporting, where qualified opinions indicate greater audit risk and can have significant adverse consequences for firms. Forecasting such qualifications has become a significant yet difficult task. Traditional econometric models such as logistic regression (LR) provide transparency but exhibit limited predictive capability, whereas modern machine learning models offer strong predictive performance but often lack interpretability. To address this gap, this study develops and evaluates an inherently interpretable predictive model of audit qualifications using the explainable boosting machine (EBM). Based on 730 firm-year observations of Indian non-financial companies listed on the Bombay Stock Exchange (BSE) during 2013–2022, the study compares EBM with LR, random forest (RF) and XGBoost. The results indicate that EBM achieves well-balanced performance across evaluation metrics, exhibits the lowest Brier score and provides stable results across classification thresholds. Its inherent interpretability enables identification of key drivers of audit qualification, while local explanations illustrate how these factors contribute differently across firms. The findings further indicate that audit qualification risk is characterized by nonlinear relationships in key financial indicators, particularly leverage and liquidity. The study contributes to audit risk literature by demonstrating how financial, governance and audit-related indicators can be integrated within an inherently interpretable machine learning framework, offering a practical decision-support tool for auditors, regulators and managers.
Shahana et al. (Thu,) studied this question.