Development and Validation of an Occupational Injury Risk Prediction Model for the Nigerian Construction Industry

Occupational injury prediction in the construction industry requires contextually validated predictive models that account for the unique regulatory, cultural, and infrastructural realities of Nigerian construction environments. This doctoral study developed and validated a machine learning-based occupational injury risk prediction model for the Nigerian construction industry. A multi-phase research design was employed. Phase One conducted a seven-year retrospective analysis of 4,200 occupational injury records from NSITF and company databases across Lagos, Abuja, and Rivers State construction firms. Phase Two collected prospective individual-level exposure data from 800 construction workers across 40 active sites over 18 months. Phase Three developed and compared predictive models using random forest, gradient boosting, logistic regression, and artificial neural network algorithms. The final model incorporated 23 predictor variables across safety climate, individual behavioral, task complexity, and environmental domains. The random forest model achieved the highest validation performance (AUC: 0.89, sensitivity: 83.4%, specificity: 87.1%). Novel predictors specific to the Nigerian context, including subcontractor status, ethnicity-based communication barriers, and proximity to insecure areas, enhanced model accuracy. The model was embedded in an accessible mobile risk assessment tool piloted at 12 construction sites. This research provides an original contribution to occupational safety informatics in low-income country construction and recommends national adoption of predictive analytics in construction safety regulation. Keywords: injury prediction, machine learning, construction safety, occupational epidemiology, Nigeria.