Predicting Project Delays Using the Novel Trended Regression Tree Method

Abstract— Accurate prediction of potential delays in pipeline projects could provide valuable information relevant for mitigating completion risk in future natural gas distribution projects. However, existing techniques for evaluating completion risk remain incapable of identifying hidden patterns in risk behavior within the huge database of projects. The purpose of this paper is to model project delays. Sample instances are drawn from the database of recent natural gas distribution projects in Iran between 2015 and 2020. A series of predictive models have been reviewed and evaluated for delay risk prediction such as k-Nearest Neighbor (k-NN) Regression, Regression Trees (RT), Support Vector Machine Regression (SVMR), and Artificial Neural Network (ANN). Computational results based on cross-validation revealed that when delays follow a rational pattern it could be predicted by our developed Trended Regression Tree (TRT) method and k-NN regression method. These novel methods are effective and provide practitioners with significantly more reliable predictions and applied insight into the delay causes. The notion of Trended Regression Trees is developed for the first time. Project delays are modeled based on project specifications and therefore there is no need to make any extra data gathering to predict project delays. Based on the research findings, we recommended that the management team focus their quest on the most effective factors to reduce project delays.