A Predictive Framework for Infrastructure Project Failure in Libya: Integrating Instability Indicators, Temporal Machine Learning, and Explainable Early Warning (Simulation-Informed Proof-of-Concept Study)

Abstract

Projects in fragile settings face political, financial, institutional, procurement, and control disruptions. This article proposes a simulation-informed framework to estimate six-month infrastructure-project failure risk in Libya using instability, financing, approval delays, organizational capacity, and project-performance signals. No real Libyan project data were available; therefore, this is a proof of concept, not a field-validated study. A synthetic longitudinal dataset (seed 20260902) covered 420 projects and 8,631 project-month records; 324 projects were used for training and 96 for temporal testing. Four classifiers were compared. In this simulation only, logistic regression achieved the highest ROC-AUC (0.700), with PR-AUC 0.381, recall 0.567, specificity 0.709, and Brier score 0.204. SHAP highlighted instability, progress gap, procurement and payment delays, financing stability, and management capacity; these are model attributions, not causal effects or evidence of Libyan field performance. The proposed Project Failure Risk Index uses normalized variables and training-derived weights. The framework offers a transparent architecture for future validation with Libyan data. Its real-world accuracy and decision value remain unknown; field validation, calibration, and governance review are prerequisites for deployment.