AI-Driven Financial Risk Early-Warning Using TD-DS-RF: A Decision Support Model Integrating Multi-Source Data

Abstract

With the swift advancement of information technology, digitalisation and intelligence have emerged as intrinsic drivers of reform and progress in enterprise risk management. Conventional approaches to financial risk early-warning predominantly rely on structured data, exhibit limited exploration of unstructured information, inadequately address the varying quality of information sources, and generally apply uniform treatment to multi-source data. In response to these shortcomings, the present study harnesses the advantages of evidence theory in managing uncertain information fusion by incorporating two-dimensional evidence theory into the financial risk early-warning framework. By integrating this with a random forest algorithm, a novel financial risk early-warning model—termed TD-DS-RF (Two-Dimensional Dempster-Shafer with Random Forest)—is established. Additionally, a financial risk early-warning lexicon is devised for publicly listed enterprises. Empirical validation utilises data from Chinese manufacturing firms listed between 2012 and 2021. The findings affirm that the TD-DS-RF model exhibits strong performance in real-world contexts, furnishing reliable decision-making support for both stakeholders and regulatory bodies, while contributing a novel conceptual lens to the domain of financial risk early-warning.