Parameter Optimization of the SVM for Big Data
Yunxiang Liu, Jiongjun Du · 2015
Wide-range traversal algorithms and some intelligent iterative algorithms are applied to the traditional SVM parameter optimization, causing much time consumption. They are not suitable for parameter optimization of big data set particularly. To get around this, a strategy of stepwise optimize parameters based on the contour plots of cross-validation accuracy is proposed in this paper. Compared with the traditional grid search algorithm, it not only shortens the time of parameters optimization remarkably, but finds a better parameter than the traditional methods based on a verification of 13910*128 data set. This paper provides an effective solution to optimize SVM parameters, especially for big data set.