Research On Vessel Collision Accident Analysis Based On Data Mining

Vessel collision accidents are the result of "people-vessel-environment-management" and other factors together.It is a multi-factor,highly-ambiguous and complex system.Analysis of vessel collision accidents by data mining has become a hotspot in water intelligent transportation.When the vessel collision accident data is analyzed and studied as the object of data mining,the main difficulty lies in the following two aspects.On the one hand,vessel collision accidents research is often based on the collision history data,and there is no open,perfect or available basic database for the vessel collision accidents.On the other hand,considering of the problem of discrete data and multi-dimension of accident data,it's very hard to find a effective analysis method.This paper is based on the establishment of the vessel collision database.By using the data mining method such as feature extraction,classification and clustering,the accident prediction,cause factor and accident distribution are analyzed.The main works are as follows:(1)The characteristics of vessel collision data are studied,and the basic database of vessel collision can be established for analysis.Vessel collision accident research is often based on the collision history data as the object of analysis,and there is no open,perfect or available historical database.Through the collection and selection of 136 vessel collision accident reports,the vessel collision data was extracted,cleaned,quantified by the use of database technology.After the structural design,it can provide data support for the collision accident analysis.(2)The analysis of vessel collision factors based on chi-square test and cluster analysis is carried out.The factor analysis is used as a feature selection problem.The 18 collision accident factors and the collision result are analyzed by chi-square test.According to the correlation between the factor and the collision result,the influencing factors are selected and divided into main and secondary factors to analyze the causes of vessel collision accident.In order to excavate the characteristic correlation model between the main factors of the accident and the collision result,the K-Modes algorithm is used to cluster the main factors.The relevant characteristics of the vessel collision are excavated according to the characteristic distribution table of the main factors.(3)This paper introduces the idea of feature extraction and classification of data mining,and proposes a prediction model of vessel collision severity analysis based on factor analysis and support vector machine for the shortcomings of existing predicting models.By using the factor analysis method,the original collision data are extracted and 10 common factors are obtained.Through the interpretation and analysis of the common factors,the implicit factors of the vessel collision accident are excavated and the importance degree of the influencing factors is sorted.Finally,the support vector machine model optimized by genetic algorithm is used to achieve the two-class and multi-class classification forecasting of the severity of vessel collision.Compared with the current popular logic regression,neural network and other prediction models,our newly proposed model has a higher classification accuracy.Compared with the prediction model constructed by the chi-square test,these two models have similar results,which proves the validity of these two methods.