<p>Earthquake damage prediction is vital to ensure occupants of buildings are not injured and substantial financial losses can be avoided. Algorithms based on machine learning are prevalent in this field. This study conducts an in depth analysis of the impact of feature selection on machine learning models for earthquake damage prediction. A feature selection method is critical for reducing data dimensionality, improving model accuracy, and mitigating overfitting. The Bat Algorithm, Particle Swarm Optimization, and Simulated Annealing were evaluated alongside twelve traditional feature selection techniques, such as Information Gain, Chi-square Test, Fisher’s Score, and LASSO Regularization. A comprehensive dataset from the 2015 Gorkha earthquake in Nepal was used to assess the predictive performance of three machine learning models-Decision Tree, Random Forest, and LightGBM. The novelty of this work lies in applying a broad spectrum of feature selection methods—many of which are being used for the first time—on the real-world dataset from the 2015 Gorkha earthquake in Nepal. A significant role is played by feature selection in the improvement of model accuracy and efficiency. Out of all the models, the Forward Feature Selection and Correlation Coefficient showed the best trade-off between processing time and accuracy. This study contributes to future disaster preparedness and structural resilience initiatives by highlighting the significance of meticulous feature selection and providing a useful framework for improving earthquake damage prediction models.</p>