基于时间序列模型SARIMA的犯罪预测研究  

Study on Crime Prediction Based on the Time-series Model SARIMA

作　　者：侯苗苗 胡啸峰[1,2] HOU Miaomiao;HU Xiaofeng(School of Information Technology and Cyber Security,People's Public Security University of China,Beijing 100038,China;Key Laboratory of Security Technology&Risk Assessment,Ministry of Public Security,Beijing 102623,China)

机构地区：[1]中国人民公安大学信息网络安全学院,北京100038 [2]安全防范技术与风险评估公安部重点实验室,北京102623

出　　处：《中国人民公安大学学报（自然科学版）》2021年第2期67-73,共7页Journal of People’s Public Security University of China(Science and Technology)

基　　金：公安部科技强警基础工作专项项目(2018GABJC01)。

摘　　要：基于时间序列分析的犯罪预测是公安情报工作的重要办法之一。利用2005年2月~2013年12月我国某北方大型城市的一般伤害、抢夺和抢劫3类犯罪案件数量数据,建立了SARIMA时间序列预测模型,并进行了验证。结果表明一般伤害案件的数量存在周期性波动,且没有明显的增减趋势,预测效果较好(PRMSE为11.95%,MAPE为10.92%)。抢夺案件的数量具有周期性波动且在2008年前后存在明显的增减趋势,通过数据处理,将抢夺数据的增减趋势与周期性分别进行了研究,得到了相对较好的预测效果(PRMSE为17.08%,MAPE为13.53%)。抢劫案件的数量不具有明显的周期性波动,难以利用SARIMA进行预测。研究结果可以应用于一般伤害和抢夺类犯罪的趋势预测,为犯罪打击提供宏观决策支持。Crime prediction based on time-series analysis plays an important role in public security intelligence analysis.In this study,a time-series model SARIMA is established and validated,using the data of crime counts of general injury,robbery and minimal violent robbery(MVR)from February 2005 to December 2013 in a mega city of the north China.The results show that the number of general injury cases has an obvious seasonal cycle but no obvious trend.The prediction performance is good with PRMSE 11.95%and MAPE 10.92%,respectively.Moreover,the number of minimal violent robbery cases also has seasonality,and there is an obvious trend of increase and decrease around 2008.Based on that,we separate the data into those for the trend and for the seasonality,respectively,and then we capture the seasonal cycle of MVR with PRMSE 17.08%and MAPE 13.53%.As for robbery,no obvious cyclical fluctuation has been found,thus SARIMA is unavailable for the prediction of this kind of crime.In general,the research results of this paper can be applied to the crime prediction of general injury and minimal violent robbery,providing macro decision support for the combat of crimes.

关 键 词：犯罪趋势预测 时间序列 SARIMA模型 一般伤害 抢夺 抢劫 

分 类 号：D917.6[政治法律—法学]