基于支持向量机的改进高斯核函数聚类算法研究

Clustering Algorithm of Improved Gauss Kernel Function Based on SVM

针对基于支持向量机的聚类算法中,由于高斯核在无限远处的衰减几乎为零,从而影响聚类效果的问题,采用了改进的高斯核函数。该方法使在高维特征空间中,核函数不仅满足在测试点附近有较快的衰减速度,而且在无限远处仍能保持适度的衰减,从而提高聚类效果。实验表明,改进的高斯核比高斯核聚类错误率更低。

Aiming at the problem that Gauss kernel in infinite distance attenuation is almost zero thus affects the clustering effect in the SVM-based clustering algorithm, an improved Gauss kernel function is adopted. The system makes kernel function not only satisfy a server decay rate in the test point nearby, but also keep modest attenuation in infinite distance for improving the clustering effect in high-dimension feature space. The experiments show that the improved Gauss kernel has a lower clustering error rate.