Enhanced model tree for quantifying output variances due to random data sampling: Productivity prediction applications

Abstract

The data used for productivity modeling represents a sample of a company's performance taken at a particular time, which leads to varying model parameters and results in a variance of the predicted productivity. This research enhances the capability of the nonlinear machine learning method called Model Tree by integrating error propagation theory to determine the variance for the predicted productivity based on a relatively large dataset available for productivity analysis. The enhanced model tree is formalized and applied in modeling structural steel fabrication productivity and piping spool fabrication productivity. The enhanced model tree is preferred over artificial neural networks due to the model's capability to explain the productivity model concerning the variance of its predicted point value and the underlying reasoning logic. The enhanced model tree will potentially find its applications in construction beyond predicting labor productivity and make a further impact on implementing explainable artificial intelligence (XAI) in wide-ranging engineering domains.