Integrating Machine Learning Into the Design of Green Building Systems

Abstract Sustainable infrastructure design is a complicated process often requiring detailed estimates specifications and constraints of the project scope to be compiled. Beyond the time-consuming gathering of project data sometimes the availability of completed projects is limited. Therefore, a method to produce similar designs with varied constraints requires a systems engineering perspective. Systems engineering provides a method to evaluate multidisciplinary design development while simultaneously following stakeholder requirements. Ecologically inspired systems have shown the ability to maintain balanced resources and structural relationships even under duress. Driven by the imperative to build sustainable infrastructure, this research explores the utilization of machine learning techniques to generate robust and reliable forecasts of green building specifications, even when design resources are scarce. To demonstrate the effectiveness of this approach, machine learning techniques were performed on a dataset of 93 green educational buildings, and on an oversampled dataset containing synthetically generated data points at the aim of certification level prediction. Both datasets contained metrics quantitatively characterizing cost, energy efficiency, and ecologically sustainable metrics specific to each building. Results indicate that the oversampled dataset allowed for better machine learning prediction among the classification algorithms considered. Oversampled data provided quality information offering cost minimization during initial design stages. This data suggests that oversampling is a reliable technique to amplify the design area of infrastructure projects when applied on data containing strong systemic classification patterns.