Singular-Value Decomposition Feature-Extraction Method for Cost-Performance Prediction: Fid-Bau Portal

Singular-Value Decomposition Feature-Extraction Method for Cost-Performance Prediction

He, Shiyuan / Du, Jing / Huang, Jianhua Z.

Earned value analysis (EVA) is a popular method used in the construction industry for cost prediction at completion. However, EVA fails to yield accurate early cost projection because of the faulty assumption that cost performance to date is static throughout the duration of a project. Recently, certain stochastic-process-based methods have been introduced to improve cost prediction performance of the classic EVA, because a project’s cost performance is manifested as a stochastic process. This paper examines a performance factor (PF) curve dimension reduction method called singular-value decomposition feature extraction (SVDFE) in order to improve the accuracy of cost projection. SVDFE predicts the trend of PF curves based on general patterns learned from all historical projects and the unique features of the particular project under prediction. SVDFE is demonstrated and validated with the case study of a real power plant project. Data analysis indicates that SVDFE significantly improves prediction accuracy and is much faster compared to other stochastic methods. Findings also suggest that SVDFE is able to capture erratic changes of cost performance throughout a project’s lifecycle and thus to provide better estimate-at-completion (EAC) predictions and early warnings.