Prediction of cost performance in construction projects using neural networks

A neural network-based program to predict construction project cost-performance is presented. The status of key variables that influence the cost of a construction project are analyzed by the program to predict the variance in project direct cost. The program consists of six neural networks, designed to predict the variances in different cost components, namely: material cost, labor cost, and equipment cost, which are later integrated into a spreadsheet program. The user is required to input the status of key variables affecting the project cost and the trained neural networks will forecast the cost variance automatically. The knowledge representation schema adopted for the development of the neural networks along with the training procedure of one of the networks is presented. The trained neural networks could successfully emulate the decision-making process of a project expert in producing revised cost estimates easily and quickly. The implementation issues of the trained networks and a discussion of the capabilities and limitations of the program conclude the paper.