Prediction of cost contingency in construction projects by introducing machine learning algorithms

Construction projects are bound by uncertainties and changes by its nature. Thus, cost contingency needs to be allocated to construction project budget to cope with any deviation of actual costs from planned ones. However, existing methods for predicting cost contingencies, as studied and practiced, still present limitations in reliability and accuracy. Machine learning (ML) has gained popularity for enhancing prediction power in various fields. The paper aims to examine various ML algorithms to implement a cost contingency prediction model, employing both continuous and categorical predictor variables. To develop the model, construction transportation project datasets, which were bid between 2013‒2017, were collected from the Florida Department of Transportation (FDOT) website. To address imbalanced regression dataset issues, the synthetic minority over-sampling technique for regression with Gaussian noise (SMOGN) algorithm is introduced. ML random forest (RF) regression associated with random search hyperparameter optimization, achieved remarkably accurate predictions compared to extreme gradient boosting (XGBoost) regression and artificial neural network (ANN) models. The results also demonstrate that four parameters are significant factors in predicting construction cost contingency: project amount, project duration, and latitude and longitude factors. These findings provide new insights for researchers in developing models and for practitioners seeking more advanced method.