Presented by Nathan Miller, Proving Ground
Predictive models using supervised machine learning have been adopted in many industries, such as banking and marketing, but are much less used within the building sciences especially to improve or inform design processes. The most significant barrier for the implementation of supervised learning in architecture is the availability of training data: supervised ML applications typically require large volumes of well-structured data to both teach themselves and to test their predictive effectiveness. While data sources are increasingly abundant in the building sciences, their effective management is limited by project complexity, fragmented processes inconsistently applied modeling standards and unsophisticated supply chains. This lecture will present recent research and projects that positions BIM data structures as an opportunity to more reliably and consistently collect facility data for use with supervised ML approaches and to impact design, construction, and operational processes.