Interpretable Artificial Intelligence (AI) Driven Public Service Delivery Monitoring Model: A Case of Zimbabwe

Abstract

This research develops an interpretable Artificial Intelligence (AI) model for public service delivery monitoring in resource-constrained governmental environments, addressing critical challenges of bureaucratic inefficiency, operational opacity, and limited technological infrastructure in Zimbabwe's public sector. Existing AI-driven governance solutions predominantly target developed economies with robust infrastructure, creating a significant gap in practical deployment models for developing contexts where lightweight architectures, explainability integration, and empirical validation remain inadequately addressed. A multi-layered neural network architecture was developed comprising data preprocessing, Natural Language Processing (NLP)-based feature extraction, ensemble machine learning for bottleneck detection, real-time performance monitoring, and Local Interpretable Model-agnostic Explanations (LIME)/SHapley Additive exPlanations (SHAP)-based interpretability interfaces. The system was empirically validated across three simulated governmental departments analyzing 1,000 administrative interactions. The model demonstrated 45% reduction in processing latency, 60% improvement in operational transparency, and 87% accuracy in bottleneck detection whilst operating efficiently within existing governmental infrastructure with minimal computational overhead. This research provides public sector administrators and policymakers with a scalable, context-aware technological solution for digital governance enhancement without substantial infrastructure investment. The study contributes a novel lightweight, interpretable AI model specifically designed for developing economy contexts, bridging the gap between AI capability and practical feasibility in resource-constrained public sector environments.