Abstract:
As earthquake casualties involved a complex of non-linear relationship between a variety of evaluation indicators, a model of support vector machine (SVM) to predict earthquake casualties based on machine learning theory was proposed. First, using principle component analysis (PCA) to make data dimension reduction for 7 influencing factors of earthquake casualties, then normalizing the extracted principal components which were used as input vectors of support vector machine, and earthquake casualties was used as output vectors, 27 typical cases of earthquake casualties were used for training data, then using grid search method (GSM)to optimize the best SVM parameters, finally the prediction model for earthquake casualties based on PCA-GSM-SVM was established, and it was used to predict the casualties of 5 samples. The result shows that the minimum error, maximum error and average error of PCA-GSM-SVM model were 5.12%,15.7% and 9.16% respectively. The average error of PCA-GSM-SVM model is reduced by 6.51% and 7.11%, respectively. Therefore, PCA-GSM-SVM model has high prediction accuracy and can be popularized in engineering practice.