基于自适应核参数优化的小波核相关向量机算法

Wavelet Kernel Relevance Vector Machine Algorithm Based on Adaptive Kernel Parameter Optimization

传统相关向量机算法在处理大规模数据集时训练速度较慢,并且高斯径向核无法完备表示特征空间。为此,基于自适应核参数优化,提出一种小波核相关向量机算法。以小波核作为基函数,在训练中,采取增量学习流程实现各个小波核参数的快速自适应优化。将提出算法应用于混沌时间序列预测及UCI数据集分类实验,结果表明,自适应参数优化小波相关向量机算法在预测精度、训练速度上均优于传统相关向量机算法。

The traditional Relevance Vector Machine（ RVM） algorithm is slow to train on large scale datasets, and the Gauss radial kernel cannot express the feature space completely. So based on adaptive kernel parameter optimization,this paper proposes a wavelet kernel relevance vector machine algorithm. Regarding the wavelet kernel as the basis function,incremental learning process is used to realize the fast adaptive optimization of each wavelet kernel parameter in training.The proposed algorithm is used on prediction of chaotic time series and classification of UCI data sets. Simulation results show that the adaptive parameter optimization wavelet correlation vector machine algorithm is superior to the traditional correlation vector machine algorithm in forcasting accuracy and training speed.