Objective: This study developed an artificial neural network (ANN) model for large-scale property valuation in England and Wales, leveraging comprehensive datasets that integrate physical, environmental, spatial, and temporal features of properties. The primary goal is to enhance the accuracy of property price prediction by capturing complex nonlinear relationships among diverse property attributes. Methods: The research utilized a Multi-Layer Perceptron (MLP) architecture with a systematic hyperparameter search to determine the optimal 10 layers configuration (512-256-128-64-32-16-8-4-2-1 nodes). Data preprocessing steps included feature encoding, standardization, and outlier removal. Spatial features, such as distances from urban infrastructure, were normalized using GIS tools. The model was trained using the Adam optimizer with mean squared error (MSE) loss and incorporated early stopping to prevent overfitting. Feature ablation studies and learning curve analyses were conducted to validate the optimal feature set (59 features) and dataset size (5.5million samples). Results: The final model achieved a coefficient of determination (R²) of 0.90, with a mean absolute percentage error (MAPE) of 15.28%, demonstrating superior performance compared to existing methods like Random Forest. The root mean square error (RMSE) was approximately 45,848 units (£), and the mean absolute error (MAE) was 29,368 units (£). These metrics highlight the model's ability to capture intricate patterns in property valuation. Conclusion: The study underscored the effectiveness of integrating deep learning techniques with spatially enhanced data for accurate mass property valuation. The MLP architecture proved particularly suited for handling heterogeneous and high-dimensional datasets, achieving state-of-the-art performance in urban environments characterized by complex spatial-economic relationships. This approach offers a robust alternative to traditional methods, enabling more precise and reliable property price predictions.
Cheraghi, F. , Mehrabi, H. and Sadeghian, A. (2026). A Deep Neural Network for Mass Property Valuation Using Spatial Features. Earth Observation and Geomatics Engineering, 9(1), -. doi: 10.22059/eoge.2026.409770.1199
MLA
Cheraghi, F. , , Mehrabi, H. , and Sadeghian, A. . "A Deep Neural Network for Mass Property Valuation Using Spatial Features", Earth Observation and Geomatics Engineering, 9, 1, 2026, -. doi: 10.22059/eoge.2026.409770.1199
HARVARD
Cheraghi, F., Mehrabi, H., Sadeghian, A. (2026). 'A Deep Neural Network for Mass Property Valuation Using Spatial Features', Earth Observation and Geomatics Engineering, 9(1), pp. -. doi: 10.22059/eoge.2026.409770.1199
CHICAGO
F. Cheraghi , H. Mehrabi and A. Sadeghian, "A Deep Neural Network for Mass Property Valuation Using Spatial Features," Earth Observation and Geomatics Engineering, 9 1 (2026): -, doi: 10.22059/eoge.2026.409770.1199
VANCOUVER
Cheraghi, F., Mehrabi, H., Sadeghian, A. A Deep Neural Network for Mass Property Valuation Using Spatial Features. Earth Observation and Geomatics Engineering, 2026; 9(1): -. doi: 10.22059/eoge.2026.409770.1199
)