Abstract: Minimum distance classifier is a simple and effective classification method. To improve its classification performance, the main methods were selecting the more effective distance measure. On the basis of analyzing the classification principle of decision tree classifier and parallelpiped classifier, a new classification method based on normalized Euclidian distance, called WMDC(weighted minimum distance classifier), was proposed. By adding weight define with nominal and string attributes and adding range restriction of attribute’s value, WMDC extended applicability of MDC(minimum distance classifier) using normalized Euclidian distance and improved its classification accuracy. Experiment results show this model has good classification accuracy in most data sets.

Key words: minimum distance classifier, Euclidian distance, parallelpiped classifier, decision tree classifier

摘要： 最小距离分类器是一种简单而有效的分类方法。为了提高最小距离分类器的分类性能,主要的改进方法是选择更有效的距离度量。通过分析多重限制分类器和决策树分类器的分类原则,提出了基于标准化欧式距离的加权最小距离分类器。该分类器通过对标称型和字符串型属性的距离的加权定义,以及增加属性值的范围约束,扩大了最小标准化欧式距离分类器的适用范围,同时提高了其分类准确率。实验结果表明,加权最小距离分类器具有较高的分类准确率。

关键词: 最小距离分类器, 欧式距离, 多重限制分类器, 决策树分类器