Data-centric AI solutions for built environment applications

This dissertation explores the application of Data-centric AI to address spatial-dependent challenges in the built environment. Spatial-dependent issues in this field require complex analysis of spatial data to interpret relationships, geometries, and locations of physical objects. Traditional methods that rely on manual or rule-based approaches are labor-intensive, slow, and limited in both accuracy and adaptability. While deep learning techniques have improved accuracy in spatial data interpretation, they often require extensive data, computational resources, and technical expertise. This dissertation introduces Data-centric AI strategies aimed at maintaining high accuracy while reducing these demands, making advanced spatial analysis more accessible. Through three case studies spanning buildings, transportation infrastructure, and industrial facilities, this research demonstrates how Data-centric AI can optimize existing datasets for improved spatial analysis and decision-making. Instead of focusing on model complexity, the study enhances data quality and utility through targeted optimization and augmentation techniques, revealing significant potential for these methods to streamline processes in urban planning and infrastructure management. This work also discusses the limitations of Data-centric approaches, such as dependency on the quality and diversity of available datasets, which can impact the robustness of outcomes in multi-class scenarios. Future research directions are suggested, focusing on expanding Data-centric AI methodologies to diverse fields, including environmental monitoring and disaster response, and encouraging broader adoption within the built environment sector. By providing a comprehensive framework for AI-driven analysis in urban development and infrastructure, this research advocates for a shift towards sustainable, data-driven approaches that foster resilience and innovation in urban planning and construction management.