恐怖组织行为挖掘与预测

发布时间：2018-03-18 03:01

本文选题：恐怖主义　切入点：全球恐怖组织数据库　出处：《北京邮电大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着全球恐怖组织势力不断壮大,全世界大部分地区都笼罩在恐怖主义阴影之下。恐怖组织攻城略地,肆意发动爆炸袭击,严重影响到人类正常生活与公共安全,其残暴行径为全世界所有爱好和平的人士所不齿,反恐斗争已经成为一项艰巨且迫在眉睫的任务。然而,以数据驱动方式对恐怖组织行为进行挖掘与预测的相关研究相对较少,而且已有的恐怖组织袭击预测算法或者预测准确率偏低,或者时间复杂度过高而不适用。本文立足于全球恐怖组织数据库GTD (马里兰大学研究人员精心收集与编排的记载了从1970年到2014年的历次恐怖袭击数据库),主要研究了恐怖组织数据库中缺失行为数据问题和恐怖组织行为预测问题,主要工作分为以下三个部分:第一,梳理了恐怖组织行为挖掘与预测的相关研究。对数据挖掘在反恐中的应用与基于数据挖掘的反恐流程进行了概括总结,对现有的恐怖组织行为预测算法进行了详细的阐明和算法分析,对本文涉及到的关联规则算法进行了概念介绍,对常用的恐怖袭击数据库进行了简要介绍。第二,构建了对恐怖袭击数据库中缺失武器数据挖掘的混合分类器,且该混合模型以朴素贝叶斯、KNN和ID3为基础模型。实验表明,通过互信息特征和随机特征进行比较,互信息特征在模型准确率上是随机特征的五倍。模型融合也比单模型有更高的预测准确率,且对武器信息预测平均F1值高达0.645。第三,构建了时序逻辑行为的有权值规则挖掘模型并提出LBT_ Weight算法。该模型将恐怖组织行为预测问题看待为NP问题,以关联规则算法思想挖掘恐怖组织行为频繁模式。通过一系列对比实验,得到一个滑动时间窗口大小为7天、滑动步长为1天、以指数函数作为衰减函数、衰减因子α= 0.22的最佳模型。LBT_ Weight算法与现有的Convexk_NN算法对比,其预测准确率比Convexk_NN高1.48%。
[Abstract]:With the growing power of the global terrorist organizations, most of the world is under the shadow of terrorism. Terrorist organizations have attacked cities and carried out indiscriminate bombings, which have seriously affected the normal life and public safety of mankind. Its brutality has been disgraced by all peace-loving people around the world, and the fight against terrorism has become a formidable and urgent task. However, there is relatively little research on mining and forecasting the behaviour of terrorist organizations in a data-driven manner. Moreover, the existing algorithms for predicting terrorist attacks or the accuracy of prediction are on the low side. Or the time complexity is too high to apply. This article is based on the global terrorist organization database GTD (University of Maryland researchers carefully collected and compiled the history of the terrorist attacks from 1970 to 2014), the main research. The problem of missing behavior data in the terrorist organization database and the problem of predicting the behavior of terrorist organization are investigated. The main work is divided into the following three parts: first, combing the related research of terrorist organization behavior mining and forecasting, summarized the application of data mining in anti-terrorism and the anti-terrorist process based on data mining. The existing terrorist organization behavior prediction algorithms are explained and analyzed in detail, the association rules algorithm involved in this paper is introduced, and the commonly used terrorist attack database is briefly introduced. A hybrid classifier for missing weapon data mining in terrorist attack database is constructed, and the hybrid model is based on naive Bayesian KNN and ID3. Mutual information features are five times more accurate than random features. Model fusion also has higher prediction accuracy than single model, and the average F1 value of weapon information prediction is 0.645. Third, In this paper, a mining model of temporal logic behavior is constructed and a LBT _ Weight algorithm is proposed, which treats the prediction of terrorist organization behavior as NP problem. The frequent patterns of terrorist organization behavior are mined by association rule algorithm. Through a series of comparative experiments, a sliding time window size of 7 days and a sliding step size of 1 day are obtained, and the exponential function is used as the attenuation function. Compared with the existing Convexk_NN algorithm, the prediction accuracy of the optimal model with attenuation factor 伪 = 0.22. LBT _ Weight is 1.48% higher than that of Convexk_NN.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：D815.5;TP311.13

【相似文献】