Predicting project progress via estimation of S-curve's key geometric feature values

• We model the position of, and the slope at, an S-curve's inflection point.
• We use six factors relating to project attributes and conditions as inputs.
• We illustrate the model using data on 51 municipal construction projects.
• We use the LM algorithm to build neural networks for input–output mapping.
• The model outperforms other models in progress prediction accuracy.

The S-curve is a commonly used tool for project planning and control that depicts a construction project's cumulative progress from start to finish. As an alternative approach to estimating S-curves, empirical models derive from progress data of past projects and use mathematical formulas to make progress a function of time. A previous study proposed a cubic polynomial for generalizing S-curves as well as a four-input neural network model for assessing the polynomial's two parameters in order to produce S-curve estimates. This paper presents an improved model, in which the two key geometric feature values of an S-curve, i.e. the position of, and the slope at, its inflection point, are used to replace the polynomial parameters as model outputs. Because these values are likely to be influenced by project conditions, two factors representing project conditions, i.e. degree of project simplicity and degree of team competence, are used as model inputs in addition to the previous four. Data on the nature and actual progress of 51 recently completed projects in the greater Kaohsiung area of Taiwan was collected to illustrate model development, in which the Levenberg–Marquardt algorithm was used to build neural networks for mapping of the input–output relationships. The new model was found to outperform other models in progress prediction accuracy for the project data collected, while sensitivity analysis confirmed its robustness.