基于核相似性的模糊多核最小二乘支持向量机

Kernel Similarity-Based Fuzzy Multi-Kernel Least Squares Support Vector Machine

最小二乘支持向量机(Least Squares Support Vector Machine, LSSVM)由于同等对待所有样例,从而易受噪声干扰,影响分类性能。模糊LSSVM的提出一定程度上克服了以上问题。本文给出了一种新的样例隶属度计算方法,其在特征空间中,利用每一样例与其他样例核相似性获得隶属度,并将其应用于模糊多核LSSVM (Fuzzy Multi-Kernel LSSVM, FMK-LSSVM),得到具有强鲁棒性的基于核相似性的模糊多核LSSVM。实验结果验证该方法的可行性与有效性。

Least squares support vector machine (LSSVM) is vulnerable to noise and affects the classification performance because it treats all samples equally. Fuzzy LSSVM overcomes the above problems by introducing membership. In this paper, we develop a new method to compute membership. In the feature space, the membership degree is obtained by using the kernel similarity between each sample and other samples, and applied to fuzzy multi-kernel LSSVM (FMK-LSSVM) to obtain a strong robust FMK-LSSVM. Experimental results verify the feasibility and effectiveness of this method.