An Original Probabilistic Risk Framework for Predicting Construction Project Failure in Nigerian Public Sector Infrastructure Delivery

Construction project failure in Nigeria's public sector, broadly defined as the non-delivery of intended scope, quality, time, or cost outcomes, is pervasive and costly, yet no predictive probabilistic risk framework calibrated to Nigerian conditions has been developed, limiting evidence-based project risk management. This study develops and validates an original Probabilistic Risk Framework for Predicting Construction Project Failure in Nigerian Public Sector Infrastructure (PRFPF-NPS), integrating probabilistic risk analysis, machine learning classification, and Bayesian network modelling. The framework is developed from empirical analysis of 240 completed and failed public sector construction projects across Nigeria's six geopolitical zones, collected through project document review, contractor interviews, and government project records over 4 years. Project success and failure outcomes were defined using a composite performance index covering cost, time, quality, and stakeholder satisfaction dimensions. A dataset of 85 project risk indicators was compiled from each project, encompassing procurement, design, site, contractor, client, political, and economic risk categories. Bayesian network modelling was used to represent causal risk relationships and generate probabilistic failure predictions from early project stage risk profiles. The framework achieves area under the ROC curve of 0.87 in cross-validated prediction of project failure category, significantly outperforming existing qualitative risk frameworks. Political interference risk and contractor financial capacity risk were identified as the two highest-impact predictors of project failure across all geopolitical zones. The PRFPF-NPS is implemented as a decision support tool for project risk assessment and recommends a systematic early warning system for Nigerian public construction project monitoring.

Keywords: construction project failure, probabilistic risk framework, Bayesian network, Nigerian public sector, predictive model