多标签分类器准确性评估方法的研究

分类是数据挖掘领域研究的核心技术之一,分类器性能评估方法也是众多学者的研究热点之一。以往的分类器性能评估方法一般针对于单标签数据集,对于多标签问题并未涉及。文中主要针对多标签分类问题中的单实例情况,提出了一种多标签分类器准确性评估方法（EMOSIML）。该方法的思路是：如果分类器对一个多标签对象预测的类别标签是其属于的多个类别标签中的任何一个,则分类结果都是正确的。该方法用C#编程实现,并对朴素贝叶斯分类器进行分类器性能评估实验,实验结果表明,EMOSIML评估方法较传统的准确率评估方法更合理。

A Study on Accuracy Evaluation Method for Multi-Label Classifier

Classification is one of the key techniques of data mining,and the classifier performance evaluation is also a hotspot.Previous classifier evaluation methods focused on single label data sets,and the multi label problem was not concerned.This article mainly aims at single instance multi label classification problem,proposes a more accurate and reasonable evaluation method(EMOSIML).The main idea of this method is that if the predict class forecasted by the classifier is one of the label set which the instanc...