Cost Prediction by Machine Learning

It is desirable to predict construction costs in the §early design stage to make sure that target costs §are met. The book investigates the possibility of §predicting the cost of construction early in the §design phase by using machine learning techniques. §Therefore, artificial neural network (ANN) and case §based reasoning (CBR) prediction models were §developed in a spreadsheet-based format. An §investigation of the impacts of weight generation §methods on the ANN and CBR models was conducted. The §performance of the ANN model was enhanced by §experimenting with the weight generation methods of §simplex optimization, back propagation training, and §genetic algorithms while the CBR model was augmented §by feature counting, gradient descent, genetic §algorithms, decision tree methods of binary-dtree, §info-top and info-dtree. Cost data belonging to the §superstructure of low-rise residential buildings §were used to test these models. Both approaches were §found to be capable of providing high prediction §accuracy. A comparison of the ANN and CBR models was §made in terms of prediction accuracy, preprocessing §effort, explanatory value, improvement potentials §and ease of use.