Application of Data Mining Techniques to Quantify the Relative Influence of Design and Installation Characteristics on Labor Productivity

The factors affecting productivity have classically been categorized as those related to the work environment and the work to be done, resulting in a piecewise understanding of productivity. In general, the factors among these categories have been considered as influencing the work environment in a mutually exclusive manner. Current industry practices of labor productivity are derived from unitized measures of piping installation under various design parameters. The heterogeneous nature of mechanical piping and plumbing projects introduce a system of installation factors that warrants simplification. This paper presents a methodological approach to develop a practical data collection metric for productivity based on established industry factors of influence. This method is developed to capture the systematic and integrative behaviors of complex piping installation factors in a simple master code structure. Although the methods applied in the paper are used to develop a productivity metric for mechanical piping, the methods could be applied to develop productivity metrics for other systems using relevant data sources. Accordingly, the paper also presents a productivity metric based on the Mechanical Contractors Association of America estimating data sources. A data mining technique utilizing a classification and regression tree (CART) algorithm is used to expose the most influential factors of piping installation on industry recognized standards of estimated labor rates without conceptual bias or industry prejudice. The optimization of progressive CART cases based on three sources of mechanical piping and plumbing estimating data results in post hoc perspectives of productivity factors that are systematically delineated and integrated across their categorical, ordinal, and scalar natures. In each case, the method provides a statistically sound and reproducible result in the form of plausible data collection metric to represent a simple industry-level coding structure capable of quantifying productivity inputs and outputs uniformly across heterogeneous piping scopes.