多元Logistic回归模型在芳腈化合物毒性等级识别中的应用

Logistic regression model in the aromatic nitrile compound toxicity level of recognition

针对芳腈衍生物对发光菌的毒性模式识别是一个非线性、高维数据处理问题,建立了多项logistic回归芳腈衍生物毒性模式识别模型。表征芳腈衍生物结构特征的指标之间存在着复杂的耦合关系,采用因子分析法对样本数据降维,消除指标间的耦合关系对识别结果的影响。选取15个芳腈衍生物的结构参数及其毒性作为建模样本数据,另8个芳腈衍生物的结构参数及其毒性作为测试样本数据,实验结果表明基于因子分析法的多项logistic回归模型对化合物毒性等级的判别正确率为95.65%,为化合物毒性模式识别提供了一种新方法、新思路。

Pick for aromatic nitrile derivatives of the luminescent bacteria toxicity pattern recognition is a nonlinear, high dimen- sion data processing problems, established the multinomial logistic regression toxic aromatic nitrile derivatives pattern recognition model. There are complicated coupling relationships between characterization of aromatic nitrile derivatives of structural features of the indicators, using factor analysis to the sampledata dimensionality reduction, eliminating the influence of coupling relationship be- tween the index on the recognition results. Select 15 aromatic nitrile derivatives structure parameters and its toxicity as the modeling sample data and the other eight aromatic nitrile derivatives of structure parameters and their toxicity as test sample data,the experi- mental results show that the multinomial logistic regression model based on factor analysis of chemical toxicity level discriminant accuracy is 95.65%,for the toxicity of compound pattern recognition provides a new method and new thinking.