Evaluation of the Accuracy of Quantity Surveying Cost Forecasts in Nigerian Infrastructure Projects Using Machine Learning

Traditional quantity surveying cost forecasting methods for infrastructure projects are frequently inaccurate in Nigeria's volatile economic environment, prompting interest in machine learning as an alternative predictive tool. This study evaluates the accuracy of machine learning models compared to traditional regression and expert estimation methods for cost forecasting in Nigerian infrastructure projects. A quantitative archival design was employed, using a dataset of 120 infrastructure project cost records from road, bridge, and public building contracts across all geopolitical zones completed between 2018 and 2024. Random forest, gradient boosting, and support vector regression models were trained and tested alongside multiple linear regression and expert estimate benchmarks. Model performance was assessed using root mean square error, mean absolute percentage error, and R-squared metrics. Findings demonstrate that gradient boosting achieves the highest accuracy with a mean absolute percentage error of 8.4% compared to 14.1% for multiple linear regression and 17.2% for expert estimates. Key predictive variables include project type, location zone, contract duration, procurement method, and contractor prequalification grade. The study provides the first systematic comparison of machine learning and traditional methods for Nigerian infrastructure cost forecasting using a geographically balanced national dataset. Limitations include data quality challenges and the black-box nature of some models. Recommendations address data governance for the quantity surveying profession and machine learning literacy in postgraduate curricula. Keywords: machine learning, cost forecasting, infrastructure projects, Nigeria, quantity surveying