【摘要】 针对传统谱算法在数据分类问题中的局限,提出一种基于共有GP-LVM和改进型SVM的数据分类算法。通过高斯过程(GP)对数据流形建立概率模型,得到高斯过程隐变量模型(GP-LVM),分析GP-LVM得到数据流形的特征信息;利用多核迭代的方式,改进SVM算法中的核函数,建立最佳的数据分类器,实现数据分类。选取FERET、UCI多类数据库进行对比实验,实验结果表明,该算法可以有效地对高维数据进行分类,针对均衡数据和不均衡数据也具有良好的分类效果,较传统算法在分类准确率上提高8%左右。更多还原

【Abstract】 The traditional spectrum algorithms had been limited in data classification problem.For its characteristics of problem,a novel algorithm based on common Gaussian process latent variable mode(CGP-LVM)and modified support vector machine(SVM)was proposed.Firstly,the probabilistic model of data manifold was established by the Gaussian process(GP),and GPLVM was gotten.The feature information was gotten by analyzing the data manifold.Secondly,the kernel function of SVM was modified by the multi-kernel iteration.Thereafter,the optimal data classifier was established.Finally,the data classification was achieved.Some data sets were selected as the experimental data,which consisted of FERET and UCI.A lot of experiments had been done.The results showed that the proposed method had not only agreat effect on accomplishing high-dimension data classification,but a great classification effect for the balanced data and the imbalanced data.The classification accuracy rate was higher than the traditional algorithms by 8%.更多还原