Author: Ali Emamifar

Publisher:

Published: 2019

Total Pages:

ISBN-13:

Change orders are an integral part of construction projects regardless of project size or complexity. Changes may cause interruption to the unchanged scope of work and working conditions and, if poorly managed, may be detrimental to project success. Many studies have been carried out to quantify the impact of change orders on construction labour productivity, with varying degrees of accuracy and variables considered. These studies reveal that quantifying loss of productivity due to change orders is not an easy task and requires a comprehensive and holistic method. There are several methods for quantifying loss of productivity, such as measured mile analysis (MMA) and the total cost method (TCM). Although measured mile analysis (MMA) is a well-known and widely accepted method for quantifying the cumulative impact of change orders on labour productivity, it is not readily applicable to many cases. In this research two models were developed to quantify losses arising from change orders. The first model does not account for the timing of change orders, but the second model considers the timing of change orders on labour productivity. Two models were developed and tested utilizing artificial neural networks and two sets of data collected by others in that field. The two datasets were statistically analyzed and preprocessed in order to transfer the data to normal distribution form and eliminate insignificant variables considered in their development. Using best subset regression, a total of seventeen variables were reduced to nine variables accordingly. Also, the study datasets were categorized into three types of timing periods; early change, normal change and late change to create the timing model. This was implemented to enable a comparison with models developed by others. Three types of artificial neural network techniques were experimented with and evaluated for possible use in the developed models. These three types are Feed Forward Neural Network, Cascade Neural Network, and Generalized Regression Neural Network. Candidate techniques were evaluated and analyzed by neural network parameters and analysis of variance (ANOVA) to select the most efficient type of neural networks, and subsequently using it to develop two models; one considers timing and the second does not. The analysis performed led to the selection of the cascade neural network for the development of the two models productivity losses due to change orders. The developed models were tested and validated utilizing several actual cases reported by others. The models were applied to a number of cases and the results were compared to those generated by frequently cited models to demonstrate their accuracy. The comparison outcome showed that the developed models can generate more accurate and satisfactory results than those of reported in previous studies.