模糊最小二乘大间隔孪生支持向量回归机

Fuzzy Twin Least Squares Large Margin Distribution Support Vector Regression

为了在不影响算法效率的前提下,将大间隔思想和最小二乘理论与模糊孪生支持向量回归机相结合,提高算法预测精度,现提出一种新的模糊最小二乘孪生大间隔支持向量回归机算法(LSFTSVR),以增加间隔分布对于训练模型的影响。理论研究证实,间隔分布在很大程度上影响模型的泛化性能。该算法在标准孪生支持向量回归机优化目标函数上增加了间隔分布的影响,同时引入最小二乘思想和模糊隶属度函数。隶属度函数基于k近邻算法计算获得每个数据点基于数据密度分布的密度加权值。提出的算法能够很好地反映训练数据集的内在分布,使数据点准确影响训练模型。在相关数据集上的实验结果表明,所提出的算法LSFTSVR比标准FTSVR算法的预测准确率更高。

In order to improve the prediction accuracy of the algorithm,the new method of least squares large interval fuzzy twin support vector regression(LSFTSVR)is proposed in order to improve the prediction accuracy of the algorithm without affecting the efficiency of the algorithm. LSFTSVR aims to increase the effect of interval distribution on the training model. Theoretical studies confirm that the interval distribution largely affects the generalization performance of the model. The algorithm increases the influence of interval distribution on the optimization of the objective function of the standard twin support vector regression,and introduces the fuzzy membership function at the same time. The membership function calculates the density weighted value of each data point based on the data density distribution based on the k-nearest neighbor theory. The proposed algorithm can reflect the inner distribution of the training data set well,and make the data point accurately affect the training model. The experimental results on related datasets show that the proposed algorithm LSFTSVR is more accurate than the standard FTSVR.