基于SVM的水上交通事故严重程度的影响因素研究

为研究水上交通事故中事故严重程度的影响因素,减小水上交通事故发生时的人员伤亡及财产损失,对2015-2016年的水上交通事故统计数据的分析.选取了水上交通事故数据中的船舶类型、事故发生时间、地点、船舶吨位、能见度和风力等级等相关因素建立了事故信息库.根据水上交通事故造成的人员伤亡数量和财产损失的大小,将事故严重程度分为3个等级,并建立了基于支持向量机(SVM)的三分类模型.然后通过交叉验证以及网格搜索算法优化SVM分类模型的惩罚参数和核函数参数,得到最优的分类模型.模型建立后,利用SVM-RFE算法求解上述影响因素对事故严重程度的权重值并排序,筛选出对于事故严重程度影响最大的因素.结果表明,支持向量机三分类模型总体分类准确率可达70% 以上;同时自沉事故、渔船事故和秋季发生的事故易造成较大的人员伤亡;危化品船舶,内河发生的事故和渔船易造成较大的财产损失. 

An Analysis of Impact Factors of Accident Severity for Water Transport Based on Supporting Vector Machine

In order to identify influencing factors for accident severity and to reduce casualties and economic loss in maritime traffic accidents,factors are extracted by developing a database of accidents information based on statistical anal-ysis of maritime accident data from 2015 to 2016.The factors mainly include ship type,accident location,time,gross tonnage of ships,visibility,and wind force,etc.According to the number of casualties and the amount of economic loss caused by maritime accidents,their consequences are divided into three levels,and a three-class model based on support vector machines(SVM)is established.Then cross-validation and a grid search algorithm are used to optimize penalty pa-rameters and kernel function parameters of the SVM model.An optimal classification model is developed.After that, SVM-RFE algorithm is used to calculate the weights of accident severity of the influencing factors.Furthermore,the fac-tors that have the greatest impacts on the consequences are identified.The results indicate that the overall classification accuracy of the three-class SVM model is larger than 70%.Self-sinking,accidents of fishing vessels,and accidents hap-pen during the autumn period are more likely to result in more casualties.Hazardous chemical ships,inland river acci-dents,and fishing vessels tend to have larger economic loss.