The potential of machine learning modeling to predict urban construction transport demand

Abstract

Purpose: Urban freight models encounter difficulties in generating construction transport demand, mainly due to a lack of knowledge on its predictors. This study investigates the potential of using data-driven approaches to predict construction site transport demand from a combination of commonly available construction project- and context-related data features. Design/methodology/approach: Machine learning (ML) models are applied to multivariate datasets, where findings show that GFA is the most important feature explaining a large part of the data variance. Findings: Using a combination of features such as GFA, project subtypes, average household income and environmental certification, the models discern enhanced data patterns. However, they struggle to predict unseen data because of the large data variance due to missing features in the dataset, differences in data sources or a large randomness in the number of transports for different construction sites. Research limitations/implications: This research underscores the importance of rigorous data collection when deploying ML for city planners and contractors, informing policy and regulations, and ultimately delivering societal gains through reduced construction transport-related disturbances. Originality/value: This study emphasizes the feature complexity influencing construction transport demand and suggests a proof-of-concept (POC) solution for future data collection.