Wang Huazhen, Lin Chengde, Yang Fan, Hu Xueqin
College of Computer Science and Technology, Huaqiao University, Xiamen 361005, PR China.
Comput Biol Med. 2009 May;39(5):425-32. doi: 10.1016/j.compbiomed.2009.02.002. Epub 2009 Apr 21.
Most classifiers output predictions for new instances without indicating how reliable they could be. Transductive confidence machine (TCM) is a novel framework that provides hedged prediction coupled with valid confidence. Many popular machine learning algorithms can be transformed into the framework of TCM, and therefore be used for producing hedged predictions. This paper incorporates random forest (RF) to propose a method named TCM-RF for classification of chronic gastritis data. Our method benefits from TCM-RF's high performance when features are noisy, highly correlated and of mixed types. The experimental results show that TCM-RF produces informative as well as effective predictions.
大多数分类器在输出新实例的预测结果时，并未表明其预测的可靠性。转导置信度机器（TCM）是一个新颖的框架，它能提供带有限定的预测并伴有有效的置信度。许多流行的机器学习算法都可以转化为TCM框架，进而用于生成带有限定的预测。本文将随机森林（RF）纳入其中，提出了一种名为TCM-RF的方法，用于慢性胃炎数据的分类。当特征存在噪声、高度相关且类型混合时，我们的方法得益于TCM-RF的高性能。实验结果表明，TCM-RF能产生信息丰富且有效的预测。
