Decision Tree Approach to Classify and Quantify Cumulative Impact of Change Orders on Productivity

Abstract

Multiple or unusual change orders often cause productivity losses through a “ripple effect” or “cumulative impact” of changes. Many courts and administrative boards recognize that there is cumulative impact above and beyond the change itself. However, determination of the impact and its cost is difficult due to the interconnected nature of construction work and the difficulty in isolating causal factors and their effects. As a result, it is very difficult for owners and contractors to agree on equitable adjustments for the cumulative impact. What is needed is a reliable method (model) to identify and quantify the loss of productivity caused by the cumulative impact of change orders. A number of studies have attempted to quantify the impact of change orders on the project costs and schedule. Many of these attempted to develop regression models to quantify the loss. However, traditional regression analysis has shortcomings in dealing with highly correlated multivariable data. Moreover, regression analysis has shown limited success when dealing with many qualitative or noisy input factors. Classification and regression tree methods have the ability to deal with these complex multifactor modeling problems. This study develops decision tree models to classify and quantify the labor productivity losses that are caused by the cumulative impact of change orders for electrical and mechanical projects. The results show that decision tree models give significantly improved results for classification and quantification compared to traditional statistical methods in the field of construction productivity data analysis, which is characterized by noisiness and uncertainty.