Framework for developing labour-hour prediction models from project design features: case study in structural steel fabrication

Abstract

Assigning labour-hours to a certain scope of work during design and estimating is still more of an art than a science. This research proposes a data-driven approach that uses multiple linear regression (MLR) and available historical data from building information models (BIM) to associate project labour-hours and project design features. The framework relies on an enhanced version of stepwise regression technique to select the most relevant predictive factors and generate a predictive model without compromising the achievable accuracy of regression. The framework also encompasses analytical methods for justifying MLR application, validating the resulting model, and establishing range estimates for point-value predictions. In collaboration with an industry partner, the framework application is exemplified by analyzing labour-hours and design features for structural steel fabrication, leading to the creation of a valid MLR model in the simplest form. Finally, pros and cons for the proposed framework and opportunities for future research are discussed.

Integrating model tree and modified stepwise regression in concrete slump prediction and steel fabrication estimating

“…The modified stepwise regression was found instrumental in identifying relevant input factors and effective in generalizing the predictive regression model. Nonetheless, as pointed out in the D r a f t conclusion by Mohsenijam and Lu (2019), "to tackle noisy, non-homogenous, and highly nonlinear data, the proposed model would likely fail due to inherent limitations of MLR." As immediate follow-up research, we integrate MT as nonlinear classifiers prior to applying MLR regressions to model complex relationships in the problem.…”

“…Many researchers had realized the challenge of having too many variables in developing a predictive model for an applied problem domain (Gardner et al 2016, Said and Prathyaj 2017, Mohsenijam and Lu 2019. Redundant input parameters in a predictive model would increase the chances of over-fitting while potentially introducing noise (Gardner et al 2016).…”

“…Unlike forward selection and backward elimination, MSR allows previously eliminated variables to be reinstated in the MLR model, while the already selected variables can be removed in future iterations. Mohsenijam and Lu (2019) proposed an application framework for developing an MLR model in its simplest form for labour-hours prediction based on design features of steel fabrication projects.…”

“…Due to the space constraint, the overview is only provided for those relevant attributes selected in the resulting model. It is worth mentioning that the same dataset was also utilized in the demonstration of a framework for developing labor-hour prediction models from project design features by applying MLR, in which input features were selected and the MLR model was streamlined by implementing MSR D r a f t (Mohsenijam and Lu, 2019). The final MLR model resulting from Mohsenijam and Lu ( 2019) is compared and cross-checked against the proposed new methodology ("M5+MSR").…”

“…Alternatively, the model tree (MT) algorithm is analogous to the merge of decision tree and multiple linear regression (MLR). An application framework was proposed by Mohsenijam and Lu (2019) for developing an MLR model in its simplest form for labour-hours prediction based on project design features. As immediate follow-up research, we integrate MT as nonlinear classifiers prior to applying MLR regressions in order to model complex relationships in the problem.…”