【摘要】 根据每个样本在整个样本空间中所处位置的分布来计算不同的权重值,可以更好地度量样本的局部性质。K近邻加权算法的有效性极大程度上依赖于参数K,但是对参数K的选取有很大的主观性。针对这一缺点,对改进的K近邻算法(dependent nearest neighbor,DNN)算法,提出一种基于D近邻点的DNN加权样本方法,并将这种新的加权方法推广到权重支持向量回归机(weighted support vector regression,WSVR)中。UCI公用数据集和股票指数数据实验结果显示,所提方法具有更高的拟合精度和更小的误差,验证了DNN-WSVR方法的可行性和有效性。更多还原

【Abstract】 Using the position distribution of each sample in the entire sample spaceto calculate the different weight values,the local nature of the sample can be better evaluated.Although the effectiveness of the K-nearest neighbor(KNN)-based weighted algorithm depends largely on the parameter K,it is very subjective to assign the parameter K with the existing methods.To tackle this problem,this paper proposes a weighted method based on dependent nearest neighbor(DNN)algorithm.Furthermore,the proposed weighted method is extended to weighted support vector regression(WSVR).The experimental results on UCI benchmark data sets and stock index data set show that the proposed method has higher fitting accuracy and smaller error,thereby verifying the feasibility and effectiveness of DNN-WSVR method.更多还原