Scholarlink

<p>The increasing reliance on artificial intelligence (AI) in financial services has raised significant concerns regarding fairness, regulatory compliance, and ethical transparency. Machine learning (ML) models used in fraud detection, credit scoring, and anti-money laundering (AML) compliance often fail to distinguish between correlation and causation, leading to biased decision-making and potential regulatory violations. Existing fairness-aware AI approaches primarily focus on post hoc bias mitigation but lack the ability to address the root causes of algorithmic discrimination. This study introduces a Causal Graph Neural Network (Causal GNN) framework designed to enhance fairness and explainability in financial AI systems by capturing the underlying causal structures of financial interactions. By integrating Structural Causal Models (SCMs) with Graph Neural Networks (GNNs), this approach disentangles causal effects from spurious correlations, thereby improving interpretability and compliance with fairness regulations such as the General Data Protection Regulation (GDPR), the Equal Credit Opportunity Act (ECOA), and Fair Lending Laws. The proposed methodology is empirically validated using real-world financial datasets, comparing Causal GNNs with traditional GNNs and fairness-aware ML models. Experimental results demonstrate that Causal GNNs significantly reduce algorithmic bias while maintaining high predictive accuracy. Furthermore, our approach improves regulatory compliance by providing interpretable causal explanations aligned with AML and Know Your Customer (KYC) policies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Causal-GNN for ethical AI in financial services: ensuring fairness, compliance, and transparency in automated decision-making

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Diego Vallarino

摘要

The increasing reliance on artificial intelligence (AI) in financial services has raised significant concerns regarding fairness, regulatory compliance, and ethical transparency. Machine learning (ML) models used in fraud detection, credit scoring, and anti-money laundering (AML) compliance often fail to distinguish between correlation and causation, leading to biased decision-making and potential regulatory violations. Existing fairness-aware AI approaches primarily focus on post hoc bias mitigation but lack the ability to address the root causes of algorithmic discrimination. This study introduces a Causal Graph Neural Network (Causal GNN) framework designed to enhance fairness and explainability in financial AI systems by capturing the underlying causal structures of financial interactions. By integrating Structural Causal Models (SCMs) with Graph Neural Networks (GNNs), this approach disentangles causal effects from spurious correlations, thereby improving interpretability and compliance with fairness regulations such as the General Data Protection Regulation (GDPR), the Equal Credit Opportunity Act (ECOA), and Fair Lending Laws. The proposed methodology is empirically validated using real-world financial datasets, comparing Causal GNNs with traditional GNNs and fairness-aware ML models. Experimental results demonstrate that Causal GNNs significantly reduce algorithmic bias while maintaining high predictive accuracy. Furthermore, our approach improves regulatory compliance by providing interpretable causal explanations aligned with AML and Know Your Customer (KYC) policies.