Seismic Risk Assessment in Educational Buildings Using Neural Networks

This study presents the implementation of Artificial Neural Networks (ANNs) for the assessment of seismic risk in educational buildings, using a public school in Lima, Peru, as a case study. Given the region’s high seismic vulnerability and the scarcity of resources for traditional structural evaluations, the proposed model integrates site-specific geotechnical, structural, and seismic variables. The ANN was structured with two hidden layers and trained through supervised learning, employing hyperparameter optimization via random search and crossvalidation to maximize predictive accuracy. The model achieved an accuracy of $\mathbf{9 2. 5 \%}$ in classifying buildings into seismic risk categories (low, medium, high), outperforming a baseline empirical vulnerability scoring method by 15%. Critical input features included building age, number of floors, construction quality, and type of structural material. Results indicate that the ANN model is a scalable, low-cost, and data-driven alternative for rapid vulnerability screening in educational infrastructure. This approach supports informed decision-making in seismic risk mitigation policies, particularly in resource-constrained contexts. Identified limitations and recommendations for extending the methodology to broader urban datasets are discussed to ensure future applicability and robustness.