快速增量加权支持向量机预测证券指数

Fast incremental weighted support vector machines for predicating stock index

传统支持向量机是对小样本提出,对于大样本会出现训练速度慢、内存占用多等问题．并且不具有增量学习性能．而常用的增量学习方法又会出现局部极小等问题．本文阐述了一种改进的支持向量机算法(快速增量加权支持向量机算法)用于证券指数预测．该算法先对指数样本做相空间重构,再分解成若干个工作子集,针对样本重要程度给出不同权重构建预测模型．实验分析表明,在泛化精度保持略好情况下,训练速度明显提高．

Traditional support vector machine （SVM） is effective only for small size of samples.When the size of sample is large, it exhibits a low training speed and a large required memory. Thus, it is not suitable for increment learning. Furthermore, traditional increment learning algorithms such as neural network have local minima only. To tackle this problem, a fast incremental weighted support vector machines for predicting the stock index is put forward. The algorithm model reconstructs the phase for the index, and then decomposes the sample space into subsets and gives different weights to them. Experimental results show that modified algorithm raises the training speed while maintaining the same precision.