What Makes a Project Safe? Identifying the Impacts Factors Have on the Safety Performance of a Construction Site through Use of Artificial Neural Networks

Abstract

What makes a construction project safe? This question prompted this research project. The goal was to identify factors and quantify their impact on the safety performance of construction projects. The first step in achieving this goal was to research key performance indicators in the area of safety and to identify common factors associated with safety in construction. A list of factors was created and presented to building construction industry members to establish causation for the factors and to eliminate any factors that did not have available data. The set of revised factors was not adequate to represent a construction project and did not fully capture the nature of their safety aspects. Safety professionals were interviewed to determine additional factors that were associated with the behavior of personnel on building sector construction sites. Historical data was collected from projects completed by a construction contractor, and this data was used to represent the revised list of factors that had been established with input from industry members. The project managers from the projects were surveyed to obtain data for the other factors identified through the interviews conducted with safety professionals. Using the historical data and information collected from surveys, a feed forward-backward artificial neural network was developed to analyze data and identify the impact that each of the factors had on safety performance. The neural network used a sigmoid transfer function with a single hidden layer. Three unique configurations of models were experimented with. Each configuration used the same data that was collected from historical project information and the surveys of project managers, as well as the same network topography; however, how the data was organized changed with each configuration. The results from each configuration had some variation but showed similar findings. The factors with the highest importance amongst all three configurations were factors that related to safety inspections and project manager mentoring.