A Shapley Value-Based Explainable AI Framework using Monte Carlo Sampling for High-Stakes Decision Making

The deployment of complex machine learning models in high-stakes domains, such as healthcare diagnostics, financial credit scoring, and criminal justice, has precipitated an urgent demand for interpretability. While deep learning and ensemble methods offer superior predictive performance, their opaque nature presents significant barriers to trust, accountability, and regulatory compliance. This paper introduces a robust Explainable AI framework grounded in cooperative game theory, specifically leveraging the Shapley value to attribute feature importance. To address the NP-hard computational complexity associated with exact Shapley value calculations in high-dimensional datasets, we propose and evaluate a Monte Carlo sampling approximation method. This approach balances computational efficiency with the rigorous axiomatic properties of the Shapley value, providing local and global explanations that are theoretically sound. We demonstrate the efficacy of this framework through extensive experimentation on financial risk assessment datasets. The results indicate that our Monte CarloThe based estimator converges rapidly to true feature contributions, offering stable and actionable insights for decision-makers operating under strict liability constraints. By quantifying the uncertainty of explanations, this research bridges the gap between black-box efficacy and the transparency required for critical societal infrastructure.