AN INTEGRATED AI AND GIS FRAMEWORK FOR TRACKING GOVERNMENT FUNDS FROM BUDGET ALLOCATION TO GROUND-LEVEL PROJECT COMPLETION

Abstract: Government development projects generate financial, procurement, physical-progress, and geospatial records at different stages of implementation. Because these records are usually maintained in separate systems, it is difficult to judge whether expenditure reported for a project is reasonably consistent with the work reported on the ground. This paper presents and evaluates an Artificial Intelligence (AI) and Geographic Information System (GIS) framework that brings these records together through a common project record. A controlled synthetic benchmark of 3,000 project records was prepared because confidential government data were not available. Fifteen percent of the records contained four deliberately injected anomaly types: financial-physical mismatch, overspending, payment delay, and procurement irregularity. Isolation Forest, Local Outlier Factor (LOF), and a rule-based baseline were compared over five random seeds using precision, recall, F1-score, ROC-AUC, PR-AUC, and Precision@K. LOF gave the highest scores in the benchmark (F1 = 0.985; ROC-AUC = 0.9999), followed by the rule-based method (F1 = 0.874; ROC-AUC = 0.982) and Isolation Forest (F1 = 0.573; ROC-AUC = 0.890). For Isolation Forest, Precision@50 was 0.680 against a 0.150 no-skill baseline. A separate 30-seed Wilcoxon signed-rank test showed a consistent difference between the rule-based method and Isolation Forest (p = 1.86 × 10⁻⁹). SHAP analysis highlighted tender variance and the financial-physical gap among the strongest contributors to Isolation Forest scores. A Moran's I test on randomly assigned coordinates found no significant spatial clustering (I = −0.016, p = 0.067). These results show that the proposed analysis pipeline can be tested in a controlled setting, while the near-perfect LOF result should be treated as a best-case outcome because the simulated anomalies were relatively easy to separate. The framework is therefore intended as an early-warning and decision-support tool, not as an automated means of establishing corruption.

Keywords: Artificial Intelligence, Geographic Information System, Government Funds, E-Governance, Anomaly Detection, Explainable AI, Public Procurement, Project Monitoring, Design Science Research