大数据背景下基于过滤式-包裹式方法的高危人员风险预警  

Risk Assessment of High-Risk Personnel Based on Filter-Wrapper Method in The Context of Big Data

作　　者：张伟 池宏 林志宏 ZHANG Wei;CHI Hong;LIN Zhihong(School of Economics and Management,UCAS,Beijing,100910;Institutes of Science and Development,Chinese Academy of Sciences,Beijing,100910;Hangzhou Hikvision digital technology Limited by Share Ltd,Hangzhou,310052)

机构地区：[1]中国科学院大学经济与管理学院,北京100910 [2]中国科学院科技战略咨询研究院,北京100910 [3]杭州海康威视数字技术股份有限公司,杭州310052

出　　处：《科技促进发展》2018年第8期742-749,共8页Science & Technology for Development

基　　金：新疆自治区公安厅2018年大数据应用项目(2018GA026):数据资源服务平台建设;负责人:王军林

摘　　要：对高危人员的犯罪风险评估是主动式警务中一项重点和核心的工作。如何基于大数据技术构建高危人员犯罪分析评估模型是其中的研究重点与难点。针对高危人员犯罪风险评估模型中的高维特征选择问题,本文设计了结合过滤式(Filter)与包裹式(Wrapper)方法的两阶段特征选择方法框架。在第一阶段Filter方法中,本文分别使用卡方检验值与KS检验值作为离散型与连续型属性的筛选指标选择了候选特征集。在第二阶段Wrapper方法中,本文设计了基于随机森林的序列后向特征选择方法进一步优选了特征集。本文使用了某地的吸毒人员数据进行了实证分析以验证方法的有效性。实验结果表明本文的方法可以有效地从高维特征集中选择出较优的特征子集,并且有较快的计算效率和良好的可解释性。The risk assessment of high-risk personnel is a key and core work in active policing. How to build a highrisk personnel crime analysis and evaluation model based on large data technology is one of the research focuses and difficulties. Aiming at the problem of high-dimensional feature selection in high-risk personnel’s crime risk assessment model, this paper designs a two-stage feature selection method framework combining Filter and Wrapper methods. In the first stage of Filter method, chi-square test value and KS test value are used as the screening index of discrete and continuous attributes to select candidate feature sets. In the second stage of Wrapper method, this paper designs a sequential backward feature selection method based on random forest to further optimize the feature set. In this paper, we use the data of drug users in some place to conduct an empirical analysis to verify the effectiveness of the method. The experimental results show that the proposed method can effectively select the best feature subset from the high-dimensional feature set,and has faster computational efficiency and good interpretability.

关 键 词：高危人员 犯罪风险评估模型 特征选择 Filter方法 Wrapper方法 

分 类 号：C93[经济管理—管理学]