Abstract:We screen 42 historical earthquake cases, and conduct principal components analysis(PCA) on seven impact factors, such as earthquake magnitude, source depth, epicenter intensity, seismic intensity, difference between epicenter intensity and seismic intensity (ΔL), population density, and occurrence moment, and construct an earthquake death toll prediction model based on particle swarm optimization(PSO) extreme learning machine(ELM). We pre-process and train the data of 37 earthquake cases, and test the accuracy of the model using the data of 5 earthquake cases. The experimental results show that the average error rate of the proposed combined PCA-PSO-ELM model is 10.87%, which is 8.70 percent points and 18.38 percent points lower than that of the PCA-ELM model and ELM model, respectively. Therefore, the combined PCA-PSO-ELM model is feasible for earthquake death toll prediction.