摘要
分类问题是机器学习领域的重要研究方向之一。支持向量机是一种基于结构风险最小化的学习机器,在解决分类问题上有着出色的效果。但基于支持向量机的分类器在处理不平衡样本时,对少类样本分类准确率偏低。诸多研究在对此问题做分析时往往把主要原因归结为各类样本间数量上的不平衡,而没有充分考虑样本点在特征空间上的分布情况。针对此问题做出原因分析,并给出结论:样本的不平衡性主要是由特征空间下各类样本的分布所决定的,而和数量上的不平衡关系较小。通过实验验证结论的科学有效性。
Classification is an important field of machine learning. SVM is a learning machine based on structur- al risk minimization, it is very good at solving classification. However its classification accuracy for the minority class of the unbalance sample set is very low. Many researchers give their analysis on it, they often consider the problem is caused by the sample unbalance in quantity. They did not consider the distribution of sample points in the feature space. The reasons are analyzed for this problem, and given the conclusion. The unbalance of classifica- tion accuracy is mainly determined by the sample distribution in the feature space, it has a smaller relationship with the imbalance in quantity. The experiment results validate conclusion.
出处
《科学技术与工程》
北大核心
2014年第3期81-85,92,共6页
Science Technology and Engineering
基金
山东省自然科学基金(2009ZRB019CE)资助
关键词
支持向量机
不平衡样本集
特征空间
样本分布
support vector machine unbalanced sample set feature space sample distribution
