Predicting the construction cost-time tradeoff using optimized hybrid deep learning for risk preference decision making

Completing projects successfully within designated cost and time constraints is a central concern of the construction industry. Earned value management is a method widely used to control project performance by estimating the cost at completion (ECAC) and estimating the schedule at completion (ESAC). Previous studies applied machine learning for these estimates, but limitations in optimization and single-objective approaches require further improvement. In this study, the optimization model of the optical microscope algorithm (OMA), neural network (NN), and bidirectional long short-term memory (BiLSTM) methods were used to develop and test a novel model, the OMA-NN-BiLSTM, for ECAC and ESAC forecasting. The conflicting nature of and interdependencies between cost and time make handling these two output variables a tradeoff problem. Thus, a multi-objective optical microscope algorithm (MOOMA) was also developed to create a Pareto curve to represent this tradeoff graphically. Additionally, an indifference curve was calculated using a triangular preference function to represent decision-maker preferences. The tradeoff results were subsequently analyzed to calculate ECAC and ESAC values based on the tangent point between the Pareto and indifference curves. For the project case period selected (4th period), the optimal tradeoff forecast indicates project completion at 574 days at an estimated cost of $2,332,958.70.