SVM in the sand-dust storm forecasting

Abstract

A novel method of the support vector machine (SVM) is proposed in the sand-dust storm-forecasting model. The development of the model includes pre-treating original data by using principal component analysis (PCA), choosing a kernel function (i.e. the Radial Basic Function (RBF) kernel), defining the search region of (C, σ2) by analyzing the influence on SVM classifier of the regularization parameter and the kernel parameter, and optimizing the two parameters (C, σ2) by using grid search in the search region. The result of the experiment shows that this SVM method has better performances than the improved back-propagation neural network (BPNN) method in terms of stability, correct classification and the running speed. © 2006 IEEE.