基于多粒度数据压缩的支持向量机

Support vector machine based on multi-granulations

将多粒度理论融合到支持向量机理论中,提出了一种基于多粒度数据压缩的支持向量机分类算法(support vector machine based on multi-granulations,MG-SVM).首先,利用多粒度理论对数据进行粒化与压缩;然后,对压缩后的数据利用支持向量机寻找最优超平面并进行分类;最后利用UCI中一些标准数据进行数据压缩与分类试验.与传统的支持向量机分类方法相比,MG-SVM算法在保持或提高经典支持向量机的分类和泛化能力的同时,有效地降低了时间复杂度.

In reality we are always faced with a large number of large scale data. Multi-granulations theory is a good tool to deal with it. Support vector machine（SVM）is a powerful instrument for solving classification problems,but it is not suitable for large-scale data. It comes into being a new idea by compromising the merits of multi-granulations theory and SVM. In this work, by introducing multi-granulations theory into SVM, a new algorithm, called support vector machine based on muhi-granulations（MG-SVM）,is proposed to deal with classification of large-scale data. First,multi granulations theory is employed to granulate data and compress data granules. Then, remove the consistent information granule to compress the data. Finally, the compressed data is used to train support vectors to find the optimal hyperplane. The experiments on some benchmark datasets show that MG-SVM algorithm not only make computational complexities decreased, but also make classification power of traditional SVM invariant.