Exploring Ensemble Learning Techniques for Ranking Project Proposals at HILTI

Abstract

Ensemble learning is a class of machine learning techniques that utilizes an ensemble of models to make a prediction. For an ensemble model to perform well it is important that the models in the ensemble are diverse, i.e. they make errors in different places. There are many different methods to achieve diverse ensembles which are being evaluated in this paper based on how well they can predict future sales of construction projects using data given in the project proposal. Furthermore, SHAP values will be used to explain the model in order to increase understanding of what the most important variables are. A new technique to determine the prediction interval for different parts of the input space for risk assessment is also presented.