基于XGBoost模型的多尺度武装冲突风险预测——以巴基斯坦为例  

Multi-scale Armed Conflict Risk Prediction Based on XGBoost Model:A Case Study of Pakistan

作　　者：王彩璇 郭黎[1] 张婉晨 马式纪 白翔天 WANG Caixuan;GUO Li;ZHANG Wanchen;MA Shiji;BAI Xiangtian(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2024年第5期545-551,共7页Journal of Information Engineering University

基　　金：河南省高等教育教学改革研究与实践重点项目(2021SJGLX299);国家重点研发计划(2021YFB3900900)。

摘　　要：全球范围内的武装冲突已经成为国际社会不容忽视的焦点,如何结合多源地理空间数据研究,从更加精细的空间尺度上预测冲突爆发风险是亟需解决的问题。以巴基斯坦为例,提出一种基于XGBoost的多尺度武装冲突预测模型。首先,构建多尺度网格对数据进行划分,经时空栅格化处理后作为预测模型的主题数据集;其次,基于数据集构建地理环境、经济社会、时间推移、空间扩张、时空依赖和全要素等6个主题子模型;最后,结合不平衡样本处理和贝叶斯自动寻优,对武装冲突爆发的风险等级进行划分和预测。基于XGBoost的多尺度网格的预测可以实现更精细的空间划分,准确捕捉冲突爆发的地点和分布模式。模型可应用于不同层级,为制定不同地区的冲突风险政策提供参考。Armed conflicts on a global scale have become the focus of the international community,and how to combine multi-source geospatial data research to predict the risk of conflict outbreak from a more refined spatial scale is a problem that needs to be solved.Taking Pakistan as an example,a multiscale armed conflict prediction model based on XGBoost is proposed.Firstly,a multi-scale grid is con⁃structed to divide the data,which is used as the thematic dataset of the prediction model after spatiotemporal rasterization.And then 6 thematic sub-models are constructed based on the dataset for geogra⁃phy,economy and society,time lapse,spatial expansion,spatio-temporal dependence and total ele⁃ments.Finally,the risk level of the outbreak of armed conflict is classified and predicted by combining the unbalanced sample processing and Bayesian auto-optimization.The XGBoost-based multiscale armed conflict prediction model can achieve finer spatial delineation and accurately capture the loca⁃tions and distribution patterns of conflict outbreaks.The model can be applied at different levels to pro⁃vide a reference for the formulation of conflict risk policies in different regions.

关 键 词：武装冲突 风险预测 XGBoost算法 多尺度网格 巴基斯坦 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]