Abstract: In order to analyze the main factors affecting the severity of waterway traffic accidents, factor analysis method was used to transform independent variables into mutually independent public factors. Then, according to the obtained main factor, the K-means clustering algorithm was used to cluster waterway traffic accident data. Finally, the Logistic model was used to establish the severity classification model for the accident data. The results show that compared with Logistic regression of potential categories, the Logistic regression model based on clustering analysis has higher accurate identification rate and prediction accuracy. Season, accident cause, ship ownership etc. are only significant in a certain category; the time period, gross tonnage and ship type etc. are significant in multiple categories. Although gross tonnage is significant in multiple categories, it has different influence directions on the severity of the accident.

Key words: waterway transportation; influencing factors; factor analysis; K-means clustering; Logistic regression

摘要： 为了研究影响水上交通事故严重程度的主要因素，利用因子分析法将自变量转化为相互独立的公共因子，根据得到的主因子，利用K均值聚类算法聚类水上交通事故数据，最后采用Logistic模型对事故数据建立严重程度分类模型。结果表明：相较于潜在类别的Logistic回归，基于聚类分析构建的Logistic回归模型准确识别率和预测精度均更高；季节、事故致因、船舶归属等仅在某一类别中显著；时间段、总吨、船舶类型等在多个类别显著，且总吨虽然在多个类别中显著，但其对于事故严重程度的影响方向不同。

关键词: 水路运输；影响因素；因子分析；K均值聚类；Logistic回归