Explainable AI (XAI) in Automated Decision-Making for E-Government: Evaluating Citizen Trust and Algorithmic Auditing Mechanisms

The integration of Automated Decision-Making (ADM) systems
powered by Artificial Intelligence (AI) in public sector governance
promises enhanced administrative efficiency, consistent welfare
allocation, and streamlined municipal licensing. However, the
deployment of “black-box” machine learning models in high-stakes
public decisions raises profound ethical, legal, and operational concerns
regarding administrative due process, algorithmic bias, and systemic
citizen distrust. This paper proposes the Public Sector Algorithmic
Auditing and Explainability Framework (PAAE-XAI), an end-to-end
architecture designed to benchmark interpretability models across
public-sector automated decision workflows. Evaluating a multi-domain
dataset of 250,000 public administrative decision cases, spanning social
welfare eligibility, automated tax compliance auditing, and municipal
building permit approvals, we establish a quantitative Citizen Trust
Index (CTI) and an Algorithmic Auditability Score (AAS). Empirical
evaluation demonstrates that implementing PAAE-XAI increases citizen
trust scores by 48.6% (elevating CTI from 0.42 to 0.89) and accelerates
independent regulatory compliance verification by 85.2% (reducing
audit latency from 14.5 hours to 2.15 hours per case). Furthermore, the
proposed multi-layered surrogate feature attribution model achieves an
explanation fidelity score of 99.1% while maintaining real-time response
SLAs under 120 ms, establishing an operational benchmark for
transparent, accountable digital governance.