偏最小二乘变量筛选法在毒品来源分析中的应用

提出了一种新的偏最小二乘变量筛选方法,该方法利用PLS回归建模过程中的一些信息,删除一部分冗余的或对建模影响不大的变量来简化、优化预报模型。用此方法结合变量扩维方法处理云南昆明、思茅、西双版纳3个来源地缴获的244个海洛因样本的ICP-MS数据时,与传统的算法比较,模型的判别准确率得到大大提高,达到95%以上。且所得到的模型含变量少,很容易分析或解释各变量对模型的影响。因此该方法可用于对毒品来源有效的识别或鉴定。

Partial Least Squares Variable Selection Method and Its Application in Drug Source Analysis

A novel variable selection method has been proposed. In this method some information in PLS modeling has been used to select original regression variables, to eliminate some unimportant or uninformative variables and to obtain the more simple and optimal prediction model. Comparing with the results from the traditional methods using this method as well as the expansion of variable dimension to deal with the data included 244 heroin samples by ICP-MS from three sources i.e. Kunming, Simao and Xishuangbanna in Yunnan province , the discrimination accuracy of model increase greatly and over 95%. Because the obtained model contains few variables, it is easy to analysis or explains the variables effect to the model. Hence the method can be used to discriminate or determine the drug sources effectively.