基于机器学习的侵财类案件危害程度分析

为了进一步分析侵财类案件的危害程度，以抢劫、抢夺和盗窃3种典型侵财类案件为例，利用ZS市2008—2014年的犯罪数据与统计年鉴数据，提取“发案时间”“发案地域”“选择时机”“选择处所”“选择对象”“人均地区生产总值”“职工月平均工资”7个特征，建立基于多种机器学习分类算法的侵财类案件危害程度预测模型，并进一步开展预测结果的分析研究。研究结果表明:梯度提升决策树（GBDT）算法性能最优，危害程度预测准确率达到了0.88；在抢劫案和抢夺案中，一般和重大的案件容易发生在繁华地带，特大案件容易发生在其他处所；侵财类案件倾向于在工作日的城区中发生，发生的危害程度大多为一般；提出的侵财类案件危害程度预测模型可为侵财类案件的风险评估及警务资源优化配置工作提供方法支持。

Analysis on hazard degree of cases of encroaching on property based on machine learning

In order to further analyze the hazard degree of the cases of encroaching on property,taking three kinds of typical cases of encroaching on property including the robbery,forcible seizure and theft as the examples,based on the crime data and statistical yearbook data of ZS city from 2008 to 2014,multiple characteristics such as “time of case incidence”,“region of case incidence”,“time of choice”,“place of choice”,“object of choice”,“per capita GDP”,“average monthly salary of employees” were extracted,then a prediction model on the hazard degree of the cases of encroaching on property based on multiple machine learning classification algorithms,and the prediction results were further analyzed.The results showed that the gradient boosting decision tree (GBDT) algorithm had the best performance,and the prediction accuracy of hazard degree reached 0.88.In the cases of robbery and forcible seizure,the general and major cases were prone to occur in the prosperous areas,and the extraordinary cases were likely to occur in other places.The cases of encroaching on property were more likely to occur in urban areas on the working days,and the hazard degrees were mostly general.The proposed prediction model on the hazard degree of the cases of encroaching on property can provide the methodological support for the risk assessment and optimal allocation of police resources on the cases of encroaching on property.