Assessment of pay item specific case-based reasoning approach for bridge unit cost estimation

A pay item specific case-based reasoning (PCBR) approach was developed and assessed for estimation of bridge repair and replacement unit costs. Rather than predict overall project or unit cost as with typical case-based reasoning (CBR) methods, the goal of the approach was to allow estimation of a collective set of critical pay item unit subcosts. Here, critical pay items are defined as those contributing 5% or more to the bridge project budget. An initial set of cost predictor variables was identified by linear regression and expert opinion, then subsequently weighted and refined individually for each subcost based on cost-predictor relationship strength. A PCBR cost model was then developed for a set of the most critical bridge repair and replacement pay items. It was used to predict future year unit costs, with cost inflation estimated with a time series projection. The effectiveness of the model was compared to a traditional nonpay item-specific CBR approach, as well as single, multiple, and Lasso linear regression models. It was found that all models provided improvements to an existing expert-based cost estimation approach in most cases. It was also found that the PCBR approach improved the traditional CBR estimate of mean cost for 74% of the pay items considered, with an average improvement of 33%. Furthermore, it reduced the coefficient of variation of 58% of the cost items, with an average reduction of 29%. Although additional computational effort is required, PCBR thus has the potential to improve cost estimates over a traditional CBR approach when pay item-specific cost data are available.