Machine-based Statistical Learning of Urban Metabolism

This dissertation aims to gain a fundamental understanding of how machine-based statistical learning (ML) can contribute and be applied to the realm of urban metabolism. Along with substantial computational advances, a deluge of ML algorithms has been successfully applied to many domains—although most applications at the time of this writing have focused on the processing of images and sounds. In the realm of urban planning and urban engineering, ML approaches have often been viewed as “mystical.” As acknowledged by many, the use of ML has received some reluctance in domains such as transportation due to their current limitations, often related to poor or lack of interpretability. In this sense, this dissertation fills knowledge gaps in the application of ML. In particular, the efforts are put into addressing interpretability (i.e., explainability), incorporating domain knowledge, and handling uncertainty to better capture patterns in resources use and human behaviors, providing more realistic conclusions in decision-making systems for urban metabolism. Furthermore, this dissertation also contributes to various domains in urban energy and resources consumption studies as these fields have not been exposed extensively to the realm of ML at the time of this writing.