基于背景审查的旅客风险分级

Classification of the passengers’risk based on their background screening

针对目前我国安检措施存在的不足,以及缺少个人背景特征研究的现状,提出基于个人背景审查的旅客风险分级方法。利用公开的恐怖分子数据集,将样本按照危险等级分为高、中、低3类,首先使用SMOTE(Synthetic Minority Oversampling Technique)方法对训练集进行过采样,之后对比朴素贝叶斯、逻辑回归、支持向量机、多层感知机和决策树几个机器学习分类方法,根据个人背景特征对旅客进行危险等级评估。数据集测试表明,使用多层感知机作为分类器准确率最高,分类准确率为80.5%,平均F1为0.80,表明能够根据个人特征对人员进行风险分级。而且在旅客分级时,应当重点审查其从事职业、出生地和社会背景3个方面。

The paper is inclined to propose a passengers’risk classification method based on the personal background screening in view of the inefficiency of the security measures at home in China in view of the lack of research on the personal background features.As is well known,John Jay&ARTIS transnational terrorism database(JJATT),a famous global terrorist database,may include over 2000 worldwide terrorists’database,such as the demographical database,the organizational institutions and the secret attacking adventures involved.Analyzing the secret data of their adventurous activities,it can be found that their sampling activities can be divided into 3 risk levels according to the places of their activities,such as Emirs,local leaders and general subordinates.Among them,Emirs and military committees are their key organizational branch members,labeled as the most dangerous branch members.And,correspondingly,their subordinated branches are known as of general risk ones,with other subordinated functional figures are known as the threatening ones in general.And,then,if,in personal risk,the classification can also be done according to the venturous degrees,the dangerous or adventurous figures,which can be identified in their timely terrorist deeds,or in accordance with their attacking manners threatening the aviation security.Since there exists a great differ-ence in the number of the 3 sampling risk levels,the training set for such figures can also be oversampled with the synthetic minority oversampling technique(SMOTE)in constructing the balanced data set.And,in addition,seeing the difference of their personal background features as the classification background,the 3 types of the risk levels as labels of classification,the training set can be classified also in different ways.For instance,comparing the classification methods of naive Bayes,logistic regression,the supporting vector machine,the multi-layer perceptron and decision tree,it would also be possible for us to find and decide suitable methods in such ways.And,finally,according to the results in choice of the testing sets,the highest accuracy rate can be chosen and determined by using the multi-layer perceptron as a classifier and achieving the classification accuracy of 80.5%with the accuracy average F1 of 0.8.Thus,the experiments can help to prove that the risk levels can be graded based on the individual characteristic features by using the decision-making tree,so as to work out the Gini indexes of the features needed for the selective features to divide the set,and,in turn,to obtain more important features in the classification.Therefore,in accordance with the results gained,when classifying passengers,attention should be focused to the passenger’s occupation,place of birth and their social background.