基于数据分类的领域自适应新算法

A novel domain adaptation approach based on data classification

一般的机器学习都假设训练数据与测试数据分布相同,而领域自适应算法则是在不同数据分布条件下进行知识传递和学习,在数据挖掘、数据校正、数据预测等领域有着广泛的应用。支持向量机SVM的主要思想是针对二分类问题,在高维空间寻找一个最优分类超平面,以保证最小的分类错误率。CCMEB理论由Tsang I提出的,是一种改进了核向量机CVM的最小包含球算法,在大样本数据集处理上有着较快的速度。而CCMEB理论同样适用于二分类的SVM数据集。将SVM理论、CCMEB理论与概率分布理论相结合,提出了一种全新的基于数据分类的领域自适应算法CCMEB-SVMDA,该算法通过计算各自分类数据组的包含球球心,能够有效地对不同领域数据进行整体校正和相似度识别,具有较好的便捷性和自适应性。在UCI数据、文本分类等数据上对该算法进行了验证,取得了较好的效果。

General machine learning assumes that the distribution of training data and test data are same,but the domain adaptation algorithms aims at handling different but similar distributions among training sets,which have a wide range of applications such as transfer learning,data mining,data correction,data projections.Support vector machine(SVM)attempts to find an optimal separating hyperplane for binary-classification problems in high-dimensional space,in order to ensure the minimum classification error rate.CCMEB proposed by I Tsang,as an improvement of the CVM,is particularly suitable for training on large datasets.In this article SVM and CCMEB are combined with probability distribution theory to formulate a novel domain adaptation approach(CCMEB-SVMDA).By calculating the center of each dataset,we can correct the dataset or identify the similarity of data between different domains.This fast algorithm has a good adaptability.As a validation we test it on the fields of'UCI data' and'text classification data'and the obtained experimental results indicate the effectiveness of the proposed algorithm.