Hybrid Gaussian Process Inference Model for Construction Management Decision Making

Construction decision-making often involves several indefinite factors, and wrong decisions usually lead to many losses and may even cause the construction to fail. Correct policy making is very important. Construction decision making used to depend on managerial staff’ experience and subjective recognition, but this approach is likely to bring about wrong decisions because of an excessive number of factors involved or biased subjective recognition. To prevent such a situation, this study establishes a Hybrid Gaussian Process Inference Model (HGPIM), which uses a Gaussian process (GP) to sort out the mapping relationship between data input and output. It also uses Bayesian inference together with particle swarm optimization (PSO) to optimize the hyper-parameters of the covariance function in GP to obtain the best inference predictive ability. By predicting with the model and giving the events that need to be decided an expected value and a variance, we can establish the data’s confidence interval as a reference for making decisions. This study collects data from three construction projects to conduct the experiment and uses HGPIM to train, predict and retest these cases to prove HGPIMs predictive ability. It also shows that the model can be applied to various cases and data and thus can be applied to construction engineering.