Development of a Predictive Model for Construction Project Cost Overrun in Nigerian Infrastructure Projects Using Machine Learning

Construction cost overruns in Nigerian infrastructure projects impose severe fiscal consequences on public budgets, yet predictive tools that can identify overrun risk at project inception are absent from Nigerian practice. This study develops a predictive machine learning model for construction project cost overruns in Nigerian infrastructure projects. A retrospective dataset of 500 completed and ongoing infrastructure projects across road, bridge, dam, and building categories will be compiled from project audit reports, World Bank procurement databases, and Bureau of Public Enterprises records covering six geopolitical zones. Predictor variables including project type, contract value, procurement method, contractor experience, design completeness, funding source, and state governance indices will be extracted. Random forest, gradient boosting, support vector machine, and logistic regression classifiers will be trained, validated, and compared using cross-validation and independent test sets. SHAP (SHapley Additive exPlanations) values will provide explainability for model predictions. This study constitutes the first machine learning predictive overrun model built on a nationally representative Nigerian infrastructure project dataset, representing an original methodological contribution to Nigerian project management science. Keywords: cost overrun prediction, machine learning, infrastructure projects, Nigeria, random forest