Artificial Neural Network-Based Prediction of Load-Bearing Capacity in Earthen Construction: A Comparative Analysis of Model Architectures In Morocco

Earthen construction has been practised for millennia due to its availability, low embodied energy, and adaptability to diverse climates. In recent years, its resurgence has been driven by sustainability concerns and the need for eco-friendly alternatives in construction. However, predicting the structural performance of earthen materials remains a challenge due to their inherent variability. This study explores the potential of Artificial Neural Networks (ANNs) to enhance the understanding and prediction of load-bearing capacity in earthen structures. A comparative analysis of different ANN architectures is conducted, examining variations in network depth, activation functions, and training algorithms. Results indicate that optimally tuned, shallow networks outperform deeper architectures by minimizing computational complexity while maintaining predictive accuracy. This work demonstrates that data-driven approaches can improve the reliability and efficiency of earthen construction, offering engineers and architects valuable tools for sustainable building design.