Abstract: The first work of distribution relief resource and improving the rescue efficiency is the survival amount prediction. The object of this paper is mainly to improve the prediction accuracy of the survival amount in earthquake disaster. First of all, the prediction indexes are put forward based on regional disaster theory and literatures. Secondly, the method of Support Vector Machine (SVM) model is introduced as the survival amount prediction in earthquake disaster to solve the index data of the small sample, high dimension and nonlinear characteristics. In this paper the model of the survival amount in earthquake disaster is put forward which replaced Mercer kernel function of inner product conditions to the Mexican mother Wavelet kernel function to effectively reduce the SVM classification of nonlinear error in high dimensional space and the limitations of conventional kernel function reducing the deviation. Finally, 53 groups of sample data are collected with the model test and these data came from the Chinese earthquake cases in 1989-2005. These sample data has the characteristics of small sample, nonlinear and high dimension that can be used to test the Mexican Wv-the SVM model. The numerical example shows Mexican Wv-the SVM model has high forecasting accuracy, fast training speed and running stability good characteristics to be compared with the standard SVM and BP neural network. In a word, the model is proved to be reliable and effective.

Key words: earthquake survival amount, prediction model, Support Vector Machine (SVM), robust loss function