Fisher正则化的最小二乘孪生支持向量机

Fisher Regularized Least Squares Twin Support Vector Machine

最小二乘孪生支持向量机是一种有效的模式分类算法,然而每一个训练样本都对最终的决策平面有影响。如果训练集含有噪声或异常点,其会过度关注这些点,这可能导致最小二乘孪生支持向量机的判别能力较差。为了解决这个问题,受Fisher准则思想的启发,本文引入了双Fisher正则化项,并在此基础上提出了Fisher正则化的最小二乘孪生支持向量机。同时,在人工数据集和UCI数据集上验证了所提算法的有效性。

The least squares twin support vector machine is an efficient algorithm for pattern classification,however each training sample has an impact on the final decision plane.If the training set contains noise or outliers,it will over-focus on these training samples,which may leads to poor discriminative ability of the least squares twin support vector machine.In or-der to deal with this problem,inspired by the idea of Fisher criterion,we introduce the least squares twin Fisher regularization terms and propose a novel Fisher regularized least squares twin support vector machine based on that for binary classification.The effectiveness of the proposed algorithm is verified on both artificial and UCI datasets.