Predictive power of ensemble learning in audit qualifications: insights from Indian firms

This study evaluates the effectiveness of ensemble learning methods – bagging, boosting, and stacking – in predicting audit qualifications among Indian firms. Traditional models often fall short because of the complexity of financial reporting data. To address these limitations, our research introduces a novel stacked model incorporating a meta-classifier and utilises a comprehensive dataset enriched with essential financial and non-financial metrics. We systematically compared the performance of seven different ensemble models with conventional classifiers. Our findings demonstrate that ensemble methods, especially our proposed stacking approach, significantly outperform standard models in key metrics such as recall, precision, and area under the curve (AUC). This study highlights the critical role of meticulous feature selection and model calibration in optimising predictive analytics. This research provides valuable insights for auditors, regulatory bodies, and policymakers, enhancing audit quality predictions and supporting financial reporting integrity in the Indian context.