A groundbreaking machine learning tool, developed through a collaboration between Italy’s Polytechnic University of Milan and Drexel University in the United States, could revolutionize how architects and urban planners predict energy consumption in neighborhoods during the early stages of design.
Published in the journal Buildings, the study introduces an artificial intelligence model capable of forecasting building energy use with 88% accuracy using just four key data points. This marks a significant improvement over existing methods, which often require extensive and detailed inputs. The research team utilized the CatBoost machine learning algorithm, which outperformed other AI models in their tests.
“This framework provides designers with quick, actionable insights into energy impacts when it matters most—during the initial planning phases,” said co-author Simi Hoque, PhD, PE, a professor of civil, architectural, and environmental engineering at Drexel University. “It empowers us to make energy-efficient decisions from the very beginning of neighborhood projects.”
Led by Andrea Giuseppe di Stefano from Polytechnic University of Milan, the team developed and validated the model using data from 22,865 buildings. The AI tool analyzes four critical factors: building size, primary use, number of floors, and climate zone. These factors were identified as the most influential through a detailed statistical analysis of the dataset.
When tested on mixed-use buildings in New York City, the model’s predictions were remarkably close to traditional energy modeling calculations, with deviations ranging from just 8.69% lower to 11.04% higher. This high level of accuracy is particularly notable given the minimal input data required.
The new approach addresses a major challenge in sustainable urban development: the need to assess energy efficiency before designs are finalized. Current energy modeling tools typically demand detailed information about building materials, mechanical systems, and operational schedules—data that is rarely available during early planning stages. In contrast, this machine learning model requires only basic information, which architects and planners usually have at the outset of a project.
The model’s success is rooted in its training on a comprehensive dataset that combines information from both residential and commercial buildings. The researchers integrated data from the Commercial Buildings Energy Consumption Survey and the Residential Energy Consumption Survey, creating a robust foundation for accurate predictions across various building types.
This study represents the first phase of a broader initiative to optimize neighborhood energy use. Future phases will explore building shapes and evaluate district-level energy systems. The team also plans to incorporate additional features, such as solar exposure analysis and district heating potential, in later stages of the project.
The research comes at a critical time, as cities worldwide face increasing pressure to reduce carbon emissions. Buildings account for approximately 40% of energy use in the European Union, and with urban populations expected to double by 2050, the demand for efficient energy planning tools is more urgent than ever.
“We’re equipping designers with practical tools to create more sustainable neighborhoods,” Hoque emphasized. “Making informed energy decisions early in the design process leads to better-performing buildings and a greener future.”
This innovative machine learning model not only simplifies energy prediction but also paves the way for smarter, more sustainable urban development.
