Junye Deng

Junye Deng

Oct. 30, 2026

Commercial real estate, Big data, Spatial layout optimization, Commercial value prediction, XGBoost, simulation.

Asset valuation and portfolio selection are both required for commercial real estate (CRE) investment, but they are often studied in isolation. This reproducible simulation links commercial value prediction to spatially constrained site selection in a stylized market of 120 sites across six submarkets, and the market reflecting stylized U.S. CRE conditions is not the same everywhere in the city. Four prediction models are evaluated by random five-fold and leave-one-submarket-out cross-validation, and their predictions are input into a binary optimisation model with capacity, budget and minimum-distance constraints. If the nonlinear value process is correctly specified, Hedonic II outperforms XGBoost in the simulated sample of 120-3000 sites (RMSE 0.052 vs. 0.083 at N=120). Compared with value-only selection, multi-criteria weighting reduces risk by 4.0% and competition by 2.5% at a 0.3% predicted-value cost, and the minimum-distance constraint achieves most of the geographic diversification. For the tested implementation and parameter budget, a genetic algorithm remains 11.2% below the certified exact solution. The results are methodological behaviours under known assumptions, not real-world performances.

© 2026, the Authors. Published by ISCCAC

This is an open access article distributed under the CC BY-NC license