论文部分内容阅读
新型犯罪日趋复杂,案件审判人员对高标准技术工作的需求不断增加,这些都要求研究新的证据评价方法,从而对各种微量物证理化检测数据的证据价值进行评估。证据评估方法能够反映法庭科学家在案件审理中的作用。这意味着上述数据(证据)应当在案件诉讼中控辩双方提出了相互对立的假设H1和H2的情况下接受评估,贝叶斯模型适用于这种情况下的证据评估。本文描述了在比较和分类(其实分类也是以比较为基础)问题中使用似然比方法(LR)对被观察的理化数据进行评估的原理。LR模型允许在一次计算中将所有重要的因素都包括在内,以此实现对相关理化数据的评估。这些因素包括,被比较样本间被比对理化数据的相似性,被测理化数据在有关总体中的稀有性,以及可能的误差来源(样本之内、之间的差异性)等。作为统计工具,LR模型只能用于仅以几个变量描述的数据库,而事实上大多数理化数据都是高度多维的(比如光谱),因此,需要使用缩维手段比如图形模型或适当的化学计量工具作降维处理,本文对此举例说明。需要指出,LR模型只应作为一种支持性(非决定性!)的工具,其结果(论)要接受严格的分析判断。换言之,统计方法并不能传达绝对的真相,采用的分析技术会有各自的不确定度,各种可能的错误答案也是统计方法的构成部分。因此,应当进行灵敏度检验亦即对所用分析方法作处理验证,从而确定其表现优劣。基于此,本文采用经验交叉熵方法举例说明如何对LR模型作校验。关于来源水平的理化数据评估,这涉及到比较样品是否源自于同一物体即是否具同一性的问题。通常,办案人员(法官、检察官或警察)会对被发现的取自于身体、衣服或鞋子的微量物证(显示与对照样本类似)是否发生了转移并留存下来的活动感兴趣,这就是所谓的活动水平检验,本文对此已有讨论分析。
The increasing complexity of new types of crime and the increasing demand for high-standard technical work by case adjudicators require the development of new evidence evaluation methods to evaluate the evidential value of various physical and chemical test data. Evidence-based assessment methods can reflect the role of forensic scientists in the hearing of cases. This means that the above data (evidence) should be evaluated in cases where the prosecution and the defense have proposed mutually hypothesis H1 and H2, and the Bayesian model applies to the evidence evaluation in this case. This article describes the principle of using likelihood ratio method (LR) to evaluate observed physical and chemical data in the comparison and classification (in fact, the classification is also based on comparison). The LR model allows all important factors to be included in a single calculation to enable the assessment of relevant physicochemical data. These factors include the similarity of the physical and chemical data to be compared between the samples being compared, the rarity of the data to be measured in the population as a whole, and the possible sources of error (within and between samples). As a statistical tool, the LR model can only be used in databases that are described in terms of only a few variables, but in fact most of the physicochemical data are highly multidimensional (such as spectra) and therefore require the use of dimensioning techniques such as graphical models or appropriate chemistry Measurement tools for dimensionality reduction, this article illustrates this. It should be noted that the LR model should only be used as a supportive (non-deterministic!) Tool, the results (theory) to be subjected to rigorous analytical judgment. In other words, statistical methods do not convey the absolute truth, the analytical techniques used have their own respective uncertainties, and all possible erroneous answers are also part of the statistical method. Therefore, the sensitivity test should be carried out that is used to verify the analysis method used to determine its performance advantages and disadvantages. Based on this, we use the empirical cross-entropy method to illustrate how to calibrate the LR model. With regard to the assessment of the physical and chemical data of the source level, this involves the question of whether the samples originate from the same object, that is, whether they have identity or not. Often, investigators (judges, prosecutors or police) are interested in activities that have been detected as metastasized and retained by trace evidence of physical presence found in bodies, clothing or shoes (similar to the control sample) Activity level test, this article has been discussed and analyzed.