After an earthquake, the rapid assessment of economic losses enables government agencies to accurately evaluate the severity of the disaster, thereby initiating the appropriate level of emergency response in a timely manner. By analysing the scope of the affected area and the scale of property losses, rescue resources can be rationally allocated to the most severely impacted regions, thereby effectively mitigating the losses caused by the disaster, while securing valuable time for emergency rescue and disaster relief efforts. To address the challenges in predicting earthquake economic losses, including numerous influencing factors, high computational demands, and complex model training, this study develops a Support Vector Machine (SVM) model optimized by Principal Component Analysis (PCA) and Genetic Algorithm (GA). PCA reduces the dimensionality of economic loss-related factors by eliminating redundancy, selecting principal components with high contribution rates as SVM inputs, with economic loss as the output. GA optimizes SVM performance parameters to establish the PCA-GA-SVM model. Testing on sample data shows it outperforms GA-SVM, GA-BP (Genetic Algorithm-optimized Back-Propagation neural network), and PCA-GA-BP models, achieving an average prediction accuracy of 95.94%, with a mean absolute percentage error (MAPE) of 4.0522%, normalized root mean square error (NRMSE) of 2.361%, and coefficient of determination (R²) of 0.9994. These results underscore the model’s accuracy and generalization ability, making it an effective tool for rapid, reliable earthquake loss prediction.