The prediction model of earthquake casuailty based on robust wavelet v-SVM

Prediction of earthquake casualty is fundamental to effectively determine the amount of emergency supply collected and to track the variation in emergency supply demand in time. Upon the lack of information for predicting casualties in the earlier stage of earthquake and due to the features of predictors as small sample and nonlinearity, this paper applies improved support vector machine (SVM) to the construction of earthquake casualty prediction model and proposes robust wavelet (RW) v-SVM earthquake casualty prediction model. Considering the disadvantage that a single loss function in SVM is not able to suppress the large amplitudes and singular points of earthquake casualty predictors, a robust loss function that allows for segmented suppression is designed by combining with Gaussian loss function, Laplace loss function and ρ-insensitive loss function, to handle different data of casualty predictors. In order to minimize the nonlinear classification error of SVM in higher-dimensional space, the independent variables in both the Morlet and Mexican parent wavelet kernel functions are replaced with a wavelet kernel function satisfying Mercer translation invariant kernel; thus, two wavelet kernel functions are obtained for machine learning, to mitigate the limitation of normal kernel function as reducing the errors. The numerical example shows that RW v-SVM model features rapid learning, high-precision prediction and advanced stability when it is used to predict earthquake casualties, which provides an effective method to solve this problem. Copyright Springer Science+Business Media Dordrecht 2015