Published January 1, 2023 | Version v1

Causal-Aware Artificial Intelligence for Explainable Financial Risk Assessment and Counterfactual Decision Support

Authors/Creators

Description

Adopting Artificial Intelligence (AI) for financial decision-making has yielded the capability to quantitatively evaluate financial risk and make data-driven decisions; however, more often than not, the machine-learning models are so complex that it\\\'s impossible to understand what drove specific predictions and act accordingly. This paper presents a Causal-Aware Artificial Intelligence for explainable financial risk assessment and counterfactual decision support. The paper presents an integrated AI framework comprised of financial-risk prediction, Explainable AI (XAI), causal analysis, and counterfactual analysis. Financial data is first pre-processed and used to train models that predict the risk conditions. Next, the XAI techniques reveal the most relevant features driving the risk prediction, while the causal analysis differentiates between predictive relationships and presumably meaningful causal relationships. Finally, the counterfactual analysis reports of the financial conditions that, if changed, could reduce the risk prediction. The AI framework is evaluated with synthetically generated financial data by comparing Logistic Regression, Random Forest, Gradient Boosting, and the proposed Causal-Aware AI models on accuracy, precision, recall, F1-score, and ROC-AUC. Feature importance, feature contribution, and counterfactual risk reduction experiments are also performed. The results suggest the potential for combining predictive models, explainability, and counterfactual reasoning to develop actionable financial risk prediction models. While the evaluation relies on synthetic data, the proposed approach is a step toward more transparent, interpretable, and actionable AI in finance.

Files

IJSET_V11_issue5_587.pdf

Files (945.2 kB)

Name Size
md5:419e49ec7f9f72ee675d59b50fa4f58c
945.2 kB Preview Download

Additional details