Early-stage Project Outcome Prediction Considering Human Factors

Abstract

In the early stages of a project, project managers need a way to connect concrete actions to the factors that affect project success. This study aims to upgrade project management methodology by using machine learning technologies to predict project results. Using a new deep learning model called “deep tensor,” we predict project results at the time of completion—including quality, cost, and delivery time—by evaluating the project’s state in its earliest stage using various types of project knowledge assets. The prediction results suggest that the predictive accuracy of the deep tensor model is more accurate than that of the random forest or multiple regression model. The way to use this model to recommend specific advice by using the factors that most influenced the model’s predictions is also presented. This research provides a method for sharing difficult-to-share knowledge across projects and will be useful for early, tangible improvement measures in the project execution phase.