Accident Modeling in Small-scale Construction Projects Based on Artificial Neural Networks

Background: Several factors contribute to accidents in small-scale construction projects (SSCPs). The present study aimed to assess the influential factors in SSCP accidents and introduce a model to predict their frequency. Methods: In total, 38 SSCPs were within the scope of this investigation. The safety index of accident frequency rate (AFR) causing 452 injury construction accidents during 12 years (2007-2018) was analyzed and modeled. Data analysis was performed based on feature selection using Pearson's χ2 coefficient and SPSS modeler, as well as the artificial neural networks (ANNs) in MATLAB software. Results: Mean AFR was estimated at 26.32 ± 14.83, and the results of both approaches revealed that individual factors, organizational factors, training factors, and risk management-related factors could predict the AFR involved in SSCPs. Conclusion: The findings of this research could be reliably applied in the decision-making regarding safety and health construction issues. Furthermore, Pearson's correlation-coefficient and ANN modeling are considered to be reliable tools for accident modeling in SSCPs.