From Black Box to Actionable Insights: An Adaptive Explainable AI Framework for Proactive Tax Risk Mitigation in Small and Medium Enterprises

Abstract

This paper proposes an adaptive explainable artificial intelligence framework designed to enable proactive tax risk prevention and control for small and medium-sized enterprises (SMEs). Traditional tax risk detection models suffer from issues such as lagging performance and black-box decision-making, making them ill-suited to dynamic tax policies and the fluctuating operations of SMEs. By integrating a policy-aware module, a few-shot incremental learning mechanism, and a dynamic feature iteration strategy, this study constructs a risk prediction system capable of real-time response to policy changes, continuous learning, and high interpretability. Experimental results demonstrate that the framework maintains high accuracy (>0.8) and AUC (>0.8) in dynamic tax environments while precisely pinpointing risk sources through interpretability techniques like SHAP. Case studies further validate its capability to provide visual risk explanations and targeted corrective recommendations in real-world scenarios. This paper charts a technical pathway for SMEs to transition from “passive auditing” to “proactive prevention” in tax compliance.