- Article
High-speed rail (HSR) expansion into secondary Southeast Asian cities raises critical questions regarding pre-operational spatial forecasting. This study develops an interpretable machine learning framework to anticipate the spatially differentiated impacts of a planned HSR station in Nakhon Ratchasima, Thailand. By integrating two decades of LandScan ambient population data, Sentinel-2 imagery, and network-based accessibility metrics, we trained an XGBoost model interpreted via SHAP analysis. To ensure methodological rigor, we implemented spatial block cross-validation, which yielded a conservative median R2 of 0.699, acknowledging the inherent spatial autocorrelation that often inflates random-split performance. The analysis reveals that recent demographic momentum dominates predictive power (mean |SHAP| = 326.53). However, HSR proximity remains a highly significant localized attractor (mean |SHAP| = 146.08). Placebo tests confirm that this HSR effect is location-specific, outperforming random locations by a factor of 3.85. Furthermore, centrality controls demonstrate that the rail effect persists independently of the existing urban core. Spatially, the station’s influence is non-stationary, with an empirical distance-decay zero-crossing identified at 5723 m. Policy scenario forecasts (2027–2037) project that a Business-as-Usual trajectory yields a modest 0.86% district-wide compound annual growth rate (CAGR). Conversely, a Transit-Oriented Development (TOD) scenario—applying a 9.08% local CAGR within the station catchment—translates to a 4.03% district-wide CAGR, potentially elevating core densities to 837.68 people/km2. These findings underscore that while HSR proximity acts as a catalyst, proactive spatial planning is the decisive factor in capturing long-term demographic dividends.
18 November 2026




![Madinah Smart City Strategy pillars and levels [21].](https://mdpi-res.com/smartcities/smartcities-09-00195/article_deploy/html/images/smartcities-09-00195-g001-550.jpg)





