摘要
采用留一交叉校验法,分别用支持向量分类法、反向传播人工神经网络和K最近邻法对72个多环芳烃类化合物致癌活性构建分类模型,并作比较.其错分样本数及预报准确率分别为:7、28和22个;90.28%、61.11%和69.44%.实验结果表明,支持向量机算法具有较强的稳健性和较好的鲁棒性,能够用于多环芳烃致癌活性的分类和预测.
In this article, support vector machine（SVM）, Artificial neural network with error back- propagation （ANN- BP） and K nearest neighbor（KNN） methods are employed to set up the model of carcinogenic properties of 72 polycyclic aromatic hydrocarbons, the results are cross - validated by the leave - one out method and compared with each other. Their wrongly classified sample number and predict accuracy are 7,28,22 and 90.28 %, 61.11%, 69.44% respectively. The experiment indicaties that SVM possess better robustness and generalization capabihty and is the best candidate to classify and predict the carcinogenic properties of polycyclic aromatic hydrocarbons.
出处
《青海师范大学学报（自然科学版）》
2005年第4期71-75,共5页
Journal of Qinghai Normal University(Natural Science Edition)
关键词
支持向量机
多环芳烃
双区理论
留一交叉校验法
support vector machine
polycyclic aromatic hydrocarbons
di - region theory
leave - one - out method
