Addressing the challenge of predicting earthquake-induced building damage, this study proposes the innovative use of Deep Neural Networks (DNN) as a solution. Focusing on optimizing predictive models, the research evaluates the effectiveness of various optimizers - ADAM, SGD, RMSprop, and Adagrad - coupled with adjustments in the learning rate to determine the most efficient configuration. The experiment was conducted to compare the performance of each optimizer in predicting post-earthquake building damage, a critical issue in disaster mitigation. The results demonstrate that ADAM significantly outperforms other optimizers, achieving the highest accuracy of up to 90.50% at a learning rate of 0.001, with RMSprop as its closest competitor. While SGD and Adagrad yielded lower accuracies, SGD showed improvement with higher learning rates. The variance analysis confirmed that the choice of optimizer significantly impacts model performance, with the p-value indicating strong statistical significance for optimizers (1.23E-09), whereas the learning rate had no significant impact (p-value 0.56098964). These findings underline the importance of selecting the appropriate optimizer to enhance the accuracy of DNN models for building damage prediction, a crucial aspect in emergency response planning and earthquake disaster mitigation efforts. This research contributes significantly to the development of more accurate predictive models, which are essential in minimizing the risks of earthquake disasters.