Conceptual Quantities Estimation Using Bootstrapped Support Vector Regression Models

Abstract

Conceptual cost models do not provide details of materials, equipment, or personnel that make up construction cost. Therefore, obtaining project details for early resource planning and cost control is difficult. A feasible solution to providing valuable details for early resource planning and cost control is to model a constituent part of construction cost, i.e., quantities. Hence, the development of models for predicting conceptual quantities of reinforced concrete structural elements is the aim of the current study. Using design parameters such as live load and soil bearing pressure, predictors were defined and used for constructing conceptual quantity models. A framework for range estimation of conceptual structural quantities using support vector regression was also presented. A total of 12 models were developed using a combination of nonparametric support vector regression and bootstrap resampling techniques. The of out-of-sample prediction intervals showed that bootstrapped support vector regression models can provide useful conceptual quantity estimates. The predicted quantities intervals can assist construction planners in early resource planning and enable performance measurement of early cost predictions throughout the construction process. This study presents two additional contributions to the existing body of knowledge apart from the proposed framework. First, an overlooked predictor variable in the literature for conceptual structural quantities, the gross soil reaction, was shown to be a good predictor of foundation quantities. Second, it shows that conceptual low-level bills of quantities estimates can be provided without sketch drawings—a necessity for measuring scope creep throughout the development of a building project.