Predicting construction schedule performance with last planner system and machine learning

Abstract

Construction project schedules deviate due to uncertainty and variability unless timely actions are implemented. While the limitations of traditional management practices help to facilitate such assessments have been widely covered, quantitative LPS research has shown empirical relations between its indicators, performance, and outcome. This paper creates a model to predict the schedule outcome during early execution using the LPS metrics and the Design Science approach. 18 solution artifacts were evaluated to predict three schedule outcome variables using 15 indicators in 1464 sample points collected from nine subsequent execution intervals from 164 projects. The selected artifact predicted the schedule outcome, which is a combination of the schedule performance at planned completion and the actual schedule deviation at completion, with a MAE of 1.24% and R2 = 0.96 averaged across the nine execution intervals, using solely LPS indicators. The model can be applied as an early warning mechanism in LPS IT-Support software.