基于折射率数据的玻璃物证比对检验似然比模型建立及评估  

Likelihood Ratio Calculation for Comparison of Glass Evidences Based on Refractive Index Data

作　　者：郭洪玲 王萍 胡灿 梅宏成 郑继利 李亚军 朱军 权养科 王桂强 GUO Hongling;WANG Ping;HU Can;MEI Hongcheng;ZHENG Jili;LI Yajun;ZHU Jun;QUAN Yangke;WANG Guiqiang(Institute of Forensic Science,Ministry of Public Security,Beijing 100038,China)

机构地区：[1]公安部鉴定中心,北京100038

出　　处：《刑事技术》2023年第4期355-363,共9页Forensic Science and Technology

基　　金：公安部技术研究计划项目(2021JSYJC08);中央级公益性科研院所基本科研业务费专项资金项目(2020JB012、2022JB009)。

摘　　要：玻璃是常见微量物证,目前国内微量物证检验报告通常只列出检验数据或定性给出比对检验结论,无法量化表达物证比对鉴定结论。似然比证据价值评价方法在全世界被广泛接受,但在中国,针对玻璃物证的定量价值评价方法还未建立,给检察官和法官准确利用玻璃物证带来困难。为提升玻璃物证的证据价值,本文拟建立量化比对检验结论的似然比方法模型并进行评估。基于四大类150个玻璃样品共750个折射率的检验数据,采用高斯核密度估计方法建立似然比(LR)模型,采用直方图和两类错误率对该模型进行评估。通过建立的似然比模型计算得出同源样品间的LR值在6.58~204500范围内,不同源的样品间的LR值在0~0.68范围内,在150个样品的自身比对中,仅有一个样品自身比对的LR<1,错误排除率为0.67%;对150个不同来源玻璃样品进行两两比对,在C^(2)_(150)=11175对比对中,有173对LR>1,错误接纳率为1.55%,两个错误率均较低,在可接受范围内。基于高斯核密度估计建立的似然比模型能够对基于玻璃折射率数据的样品比对给出满意的比对结果,可为法庭提供玻璃样品比对的量化评估值。Glass is frequently encountered trace evidence in forensic investigations.Currently,the examination reports of glass can only present the physicochemical data or give comparison results qualitatively,lacking a quantitative evaluation index to draw conclusions.The likelihood ratio(LR)approach has been widely used in forensics to provide a measure of evidence value to evaluate the contribution of evidential physicochemical data all around the world.But in China,the quantitative evidence evaluation method has not yet been built yet for glass evidence,which poses judges and prosecutors diffi culties to use glass evidence accurately in the trial.In order to provide a quantitative interpretation and evaluation of the evidential value of glass,an LR model was tried to be built and its performance was evaluated.A kernel density estimation using a Gaussian kernel was carried out to estimate the probability density function for 750 refractive index data of 150 different glass samples.An LR model was built based on the probability density function.Moreover,the histogram of log LR and false inclusion and false exclusion rates were used to evaluate the LR model using LR=1 as the threshold.The result showed the LR values of within-source pairwise comparisons were in the range of 6.58 to 204500,and the between-source range is 0 to 0.68.Evaluation of the results revealed low misleading rates.Only one pair within-source comparison was found with LR<1(expected to be>1).The false exclusion misleading rate was 0.67%.There were 173 pairs of between-source comparisons among pairs were found with LR>1(expected to be<1),leading to 1.55%false inclusion misleading rate.The likelihood ratio model built based on Gaussian kernel density estimation presents a satisfactory result for glass sample comparisons based on glass refractive index data,providing a practicable quantitative evaluation method for comparing glass samples.

关 键 词：法庭科学 玻璃 折射率 似然比 核密度估计 错误率 错误排除率(FE) 错误接纳率(FI) 

分 类 号：DF794.3[政治法律—诉讼法学]