Answers are grounded in this paper and cite their sources.
Accurately forecasting residential energy demand during conceptual architectural design is vital for building decarbonization, yet early models often struggle with balancing input complexity and predictive speed. In this work, we developed a machine learning-driven decision-support framework evaluating five regression architectures Random Forest (RF), Artificial Neural Networks (ANN), Decision Trees (DT), Support Vector Regression (SVR), and Multiple Linear Regression (MLR) across a dataset tailored to Mediterranean residential structures. Eleven architectural and environmental variables were initially evaluated before applying dimensionality reduction to examine how parameter depth affects overall model precision. The results directly challenge the conventional assumption that richer parameter sets automatically improve performance. Rather, reducing input dimensionality down to four decisive parameters floor area, cooling demand, climate classification, and envelope insulation quality minimized data noise and maximized generalization capability, reaching high accuracy (R²> 0.95). Furthermore, introducing vertical spatial analysis showed that story height substantially alters energy behavior: intermediate floors function as passive thermal buffers with stable, minimal loads, whereas envelope-adjacent ground and roof levels exhibit heightened thermal variance and elevated demand. Consequently, embedding vertical floor orientation alongside condensed key parameters establishes a lean, highly reliable protocol for early-stage energy evaluation. Overall, the proposed framework demonstrates that reliable and high-precision energy predictions can be achieved using a limited set of performance-oriented design parameters. The proposed approach provides architects and designers with an efficient, data-driven tool for evaluating energy performance without relying on computationally intensive simulations during the early design process.
No takes yet. Share an insight, caveat, or question.
Setenay UÇAR (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: