Enhancing Going Concern Prediction With Anchor Explainable AI and Attention-Weighted XGBoost

In the rapidly evolving sector of financial analytics, predicting a firm’s going concern status accurately is vital for informed user decisions. This study introduces a novel method that synergizes Anchor Explainable Artificial Intelligence (XAI) with an Attention-Weighted Extreme Gradient Boosting (XGBoost) model, significantly improving the precision and clarity of going concern predictions. Traditional models often trade off explainability against complexity, diminishing user confidence. Our solution, integrating Anchor XAI, offers lucid, comprehensible explanations for the model’s predictions, enhancing trust and interpretability. The developed Attention-Weighted XGBoost algorithm, targeting essential financial indicators, markedly surpasses traditional approaches in prediction by its 98% accuracy, as evidenced by improved precision and recall. This integration not only makes the prediction process more transparent but also advances the field towards more interpretable AI solutions. Additionally, our approach highlights important features specific to each class, distinguishing our findings with significant indicators like working capital/total assets, and total equity for potential non-going concerns, debt ratio, current assets to total liabilities ratio, and long-term funds to fixed assets ratio for going concerns, alongside governance factors such as the number of independent directors and BIG4 audit status for entities with going concern doubt. These advancements demonstrate the effectiveness of melding explainable AI with attention mechanisms to bolster the trustworthiness and clarity of financial forecasts, opening new research paths in financial analytics.