Embodying learning effect in performance prediction

Predicting performance of contractors is of interest to both academics and practitioners. The physical execution of a project is critical to the overall success of the development. Having a competent contractor that can deliver is most desirable. In this aspect, a significant number of performance prediction models have been developed. Multiple regression and neural networks are typically used as the analytical tools in these prediction models. This paper reports a study that employs a learning curve approach to perform the prediction task. It is suggested that this approach can accommodate the changes in performance as experience accumulates. Thus a performance pattern is projected in addition to the project final outcome. A two-step approach suggested by Everett and Farghal was adopted for this study. First, the learning curve model that best represents a contractors' performance was explored using the least-square curve fitting analysis. Second, prediction analysis was performed by comparing the actual performance data with their respective prediction results obtained from extrapolation on the selected learning curve. The three-parameter hyperbolic model was found to provide the most reliable prediction on performance in this study.