ORCID

0009-0004-1553-5689

Keywords

Building Information Modeling (BIM), Machine Learning (ML), Random Forest, Energy Prediction, Carbon Prediction, LEED Certification.

Abstract

This research developed and validated a Building Information Modeling (BIM)-based Machine Learning (ML) framework for predicting the energy and carbon performance of office buildings during early design stages. The study addresses the need for reproducible data-driven methods supporting sustainable design decisions with reduced simulations time. The proposed approach integrates Revit and Insight with statistical modeling in Weka, creating an automated, transparent, and regionally adaptable workflow for energy and carbon prediction from a BIM-generated data. A reduced-factorial Design of Experiments (DOE) guided the generation of 260 parametric Insight simulations, including base, generalization, and stress-test models distributed across six U.S. climate zones. Each model varied geometric, envelope, system, and operational parameters, forming a comprehensive dataset of 14 independent variables and five dependent metrics: Energy Use Intensity (EUI), Operational Energy (OE), Operational Carbon (OC), Embodied Carbon (EC), and Total Carbon (TC). Four regression algorithms - Linear Regression (LR), M5P, SMOReg, and Random Forest (RF) - were trained and evaluated using 10-fold cross-validation. The RF model achieved the highest overall accuracy, with R² > 0.97 and mean absolute errors below 5 % across all metrics. Feature importance analysis identified HVAC system type, window-to-wall ratio, and operation schedule as the most influential variables. Validation with two LEED v3 Gold-certified University of Central Florida buildings confirmed that the RF-based surrogate accurately reproduced measured energy trends and, after applying regional carbon-factor corrections (0.331 kg CO₂/kWh, EPA eGRID 2023), achieved improved calibration between simulated and measured carbon intensities, reproducing proportional operational trends across both validation buildings. The results demonstrate that BIM-ML integration enables fast, reliable performance estimation, bridging simulation and artificial intelligence to support informed, energy-efficient and low-carbon design decisions during the earliest project stages.

Completion Date

2025

Semester

Fall

Committee Chair

Amr Oloufa

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Department of Civil, Environmental and Construction Engineering

Format

PDF

Identifier

DP0029730

Document Type

Thesis

Campus Location

Orlando (Main) Campus

Download

Share

COinS