基于自适应距离度量的分类器设计方法

Classifier design based on adaptive distance metric

通过对欧氏距离度量的分析,提出了自适应距离度量。首先利用训练样本建立自适应距离度量模型,该模型保证了训练样本到相同模式类的距离最近,到不同模式类的距离最远,根据该模型建立目标函数,求解目标函数,得到最优权重。基于最小距离分类器和K近邻分类器,采用UCI标准数据库中部分数据,对提出的自适应距离度量和欧氏距离度量进行了实验比较,实验结果表明自适应距离度量更有效。

After the Euclidian distance metric is analyzed, a new training method based on adaptive distance metric is proposed, which adopts a model about adaptive distance metric, the model assures that the distance between training sample and the same pattern classification is nearest, and the distance between training sample and other pattern classification is far, then a optimal Weight is obtained through solving objective function. By adding weight def＇me for distance in the classification phase, the classifier improved its classification accuracy. Experiment is tested on UCI standard database, the results show that the proposed minimum distance classifier and Knearest neighbor classifier based on adaptive distance metric is effective, and it is superior to Euclidian distance metric in classification performance.