ELMs和SVMs在多分类问题上的泛化性能比较

多分类问题是机器学习、数据挖掘领域的重要研究内容。在文本分类、语音识别、图像识别、基因检测等方面有广泛的应用。通过在UCI数据集对极限学习机算法ELMs(ELM,KELM)和支持向量机算法SVMs(SVM,LSSVM)在多分类问题上的表现进行详细比较,得出以下结论:ELMs相较于SVM在多分类问题上有更高的分类准确率,而且随着分类数目的增加,ELMs的泛化能力相较于SVM提高越多,但是ELMs对于LSSVM并没有得到上述结论;ELMs相较于SVMs对数据的类别数目不敏感,分类准确率随类别数目增加下降不明显;ELMs相较于SVMs在多分类问题上所需计算代价更小,且拥有更快的学习和训练速度,适用于多分类问题。

COMPARING ELMS AND SVMS GENERALIZATION PERFORMANCE ON MULTI-CLASS CLASSIFICATION PROBLEM

Multi-class classification problem is an important issue in machine learning and data mining.It is widely used in text classification,speech recognition,image recognition and gene detection.In this paper,we compared the performance of ELMs(ELM,KELM) and SVMs(SVM,LSSVM) on multi-class classification problem in detail through UCI dataset.We draw the following conclusion: compared with SVM,ELMs has higher classification accuracy on multi-class classification problem,and with the increase of classification number,the generalization ability of ELMs is more improved than that of SVM,but ELMs does not get the above conclusion for LSSVM;compared with SVMs,ELMs is insensitive to the number of classes,and the classification accuracy decreases little with the increase of the number of classes;compared with SVMs,ELMs has less computational cost,and has faster learning and training speed,which is suitable for multi-class classification problem.