Abstract:
Accurate prediction of construction schedule and cost plays critical role to project success. Many quantitative and associative models have been developed for more accurate prediction. However, these models often lack robustness due to bias and variance. Ensemble type of machine learning algorithm can perform well for prediction by balancing bias and variance.
This study aims to develop construction schedule and cost prediction modelusing one of the recent ensemble machine learning algorithms named Gradient Boosted Regression Tree (GBRT).Data were obtained from 69 construction projects of Dhaka city of Bangladesh. These projects were categorized as low rise, medium rise and high rise buildings according to the number of floors. One-way ANOVA F-test has been applied to select the statistically significant features. Finally, the regularized GBRT has been applied to develop the construction schedule and cost prediction models. Performances of regularized GBRT models were compared to Support Vector Regression (SVR) and Multiple Linear Regression (MLR) models. Mean absolute percentage error (MAPE) and mean squared error (MSE) were used as performance metrics.One-way ANOVA feature selection method reveals that location, land size, floor height, floor area, number of basement, workforce level and number of floor had significant impact on schedule and cost prediction model for low rise buildings. For medium and high rise buildings,land size, floor area, number of basement, workforce level and number of floor are the most significant features. The resultsshow that regularized GBRT models havelower MAPEs and MSEs than SVR and MLR models. Therefore, regularized GBRT models have performed better than SVR and MLR models in construction schedule and cost prediction for low, medium and high rise buildings.
