基于PCA和M-SVMs的化学物质生态危害预测应用研究

为防止新化学物质投入市场时对生态环境造成危害,需对其生态危害程度进行评价。现有评价方法把各指标对生态危害的贡献看成是等效的,不能客观反映事实,且评价指标较多,指标之间具有较强的相关性,会降低预测精确度,为了解决该问题,文章将主成分分析和支持向量机相结合。首先运用主成分分析进行特征提取,降低数据维数,获取数据的主要信息;然后将二值分类支持向量机扩展到多类支持向量机,利用多类支持向量机建立化学物质生态危害预测模型,采用10折交叉验证法对模型进行检验,得到平均正确率达到89.24%。并与未进行主成分分析的支持向量机分类模型进行了比较,实验结果表明该方法具有更好的预测精度,值得推广。

Application Research for Predicting Ecological Risk of Chemicals Based on PCA and M-SVMs

When evaluating new commercial chemicals regarding their potential ecological risks,it appeared that the existing methods,lacked adequate abilities to objectively reflect the risks,which might result from the fact that there often existed too many factors that were closely correlated.Two methodologies were introduced,i.e.the principal component analysis(PCA) and the support vector machine(SVM),and using PCA combined with the binary SVM that was extended into the multi-class SVMs(M-SVMs),a new prediction model of chemicals’ eco-risk was then established.Thereafter,by use of M-SVMs and 10-fold cross validation method the model was verified,which suggested a rather high average accuracy rate of 89.24%,and its better prediction character was proved as well by the comparative study which was made between the new model and another SVM classification model without using full factors.