Audited financial statements are central to investor trust, and audit opinions serve as critical signals of reporting quality. This study leverages automated machine learning (AutoML) to predict qualified versus unqualified opinions for Indian listed firms using a balanced dataset of 730 firm-year observations (2013–2022). The analysis employs H2O AutoML with stacked ensembles, gradient boosting, random forests, XGBoost, and deep learning models. The best-performing ensemble achieved an AUC of 0.93 and AUCPR of 0.94, demonstrating excellent discriminatory ability. Threshold calibration further improved accuracy and F1-score by reducing false positives without compromising sensitivity. Interpretability methods, including SHAP values and partial dependence plots, identified efficiency (asset turnover) and leverage (total debt to total assets) as the most influential drivers, consistent with Agency Theory and the risk-based auditing framework. The study contributes by demonstrating the utility of AutoML in audit analytics, the value of threshold tuning, and the role of interpretable AI in bridging theory and practice.

Keywords

Audit opinion, AutoML, Interpretable AI, Predictive Analytics