Bankruptcy prediction: Integration of convolutional neural networks and explainable artificial intelligence techniques

Abstract

Accurately predicting a company’s future performance is vital for both management and investors. This study employs the Explainable Artificial Intelligence (XAI) approach, utilizing a Convolutional Neural Network model (CNN) to forecast company financial conditions based on their financial ratios. By transforming financial data into images, we introduce a Bankruptcy Prediction Model (BPM) that enhances interpretability through techniques like Shapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME). These XAI methods aim to clarify AI decisions in the BPM, addressing the ongoing debate within the financial research community regarding the most informative ratios for bankruptcy prediction. This research marks a significant advancement in financial accounting by merging the transparency of XAI with effective bankruptcy prediction, offering a more comprehensive understanding of financial ratio analysis.