Multi-scale relevance vector machine classification based on intelligent optimization

Abstract

An appropriate selection of kernel function and its parameters is very important for the relevance vector machine (RVM) to achieve a good performance. To overcome the limitation of RVM with single kernel, a multi-scale RVM classification method based on intelligent optimization is proposed. Multiple Gaussian kernels are combined by linear weighting and the kernel parameters are tuned by quantum-behaved particle swarm optimization (QPSO) algorithm. The experimental results show that the proposed method has higher classification accuracy than typical RVM classifiers with single kernel.