Reliable price estimation supports housing policy and market transparency in Saudi Arabia, yet the online listings that form the main public source of price data are noisy, heavily duplicated, and describe many attributes only in Arabic free text. This study develops and evaluates a reproducible pipeline for predicting apartment prices from 19,971 for-sale listings scraped from the Aqar.fm platform. A documented cleaning procedure removed 25.2% of the raw records, chiefly repeated advertisements, and retained 14,934 listings concentrated in Jeddah, Riyadh, and Dammam. Fifty binary features were derived from the Arabic descriptions with spelling-variant-aware keyword rules. Multiple linear regression, ridge regression, random forest, LightGBM, and XGBoost were compared on log price under five-fold grouped cross-validation, which keeps listings sharing a description group out of the opposite partition, with hyperparameters selected over widened grids. LightGBM was selected by the cross-validation criterion and outperformed the linear baselines on every test-set metric at both the log-price and price scale: a mean absolute error of 69,985 SAR (95% bootstrap CI 65,005–75,683) and MAPE of 9.3%, against 93,161 SAR and 12.9% for multiple regression. Location features accounted for most of the gain over structural attributes alone (MAE 134,218 to 73,645 SAR), Arabic text describing the building added a further 4.9% reduction (to 70,016 SAR), and neighbourhood-service text added no further reduction in error. A model trained on only two of the three cities performed poorly when tested on the held-out city (MAPE 20.2% versus 7.0% under a random split; R² −0.28), a result attributable mainly to location-identity encodings that are undefined for an unseen city rather than to a general failure of transfer. Permutation importance, grouped permutation importance, and TreeSHAP analyses are reported to explain the selected model's predictions, which are interpreted as predictive associations rather than causal effects.
Baqasah, A. M. (2026). Predicting Apartment Prices from Arabic Online Listings in Saudi Arabia: Gradient Boosting with Structural, Locational and Text-Derived Features. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/IJACSA.2026.0171044
Baqasah, Abdullah M.. "Predicting Apartment Prices from Arabic Online Listings in Saudi Arabia: Gradient Boosting with Structural, Locational and Text-Derived Features." International Journal of Advanced Computer Science and Applications, vol. 17, no. 10, 2026, https://doi.org/10.14569/IJACSA.2026.0171044.
@article{Baqasah2026,
title = {Predicting Apartment Prices from Arabic Online Listings in Saudi Arabia: Gradient Boosting with Structural, Locational and Text-Derived Features},
journal = {International Journal of Advanced Computer Science and Applications},
volume = {17},
number = {10},
year = {2026},
publisher = {The Science and Information Organization},
author = {Abdullah M. Baqasah},
doi = {10.14569/IJACSA.2026.0171044},
url = {https://doi.org/10.14569/IJACSA.2026.0171044}
}
Open Access — licensed under a
Creative Commons Attribution 4.0 International License.
Unrestricted use, distribution, and reproduction in any medium, even commercially, as long as the original work is properly cited.