Evaluation Of Accident Black Spots On Roads Using Clustering Algorithm

The solution of controlling traffic accident is the important part of traffic safety management. It has effect on the advantage of whole country and all the people directly. Along with the development of computer technology, it become realism to deal with road traffic accident using artificial intelligence,expert system,data mining and so on. These technologies show more advantage nowadays. Consequently the Intelligent Transportation System has been put forward. ITS is a newly technical system with the steps which is composed of technology development and industrialization. It can solve the traffic problem using some technology, for example information technology,computer technology ,organizational technology and so on. With the development of traffic project at very fast speed, the increasingly traffic information gradually become valuable wealth for ITS. The traditional modeling and classical mathematic formula can not content the request of information analysis, because the information's characteristic, for example widely source,various kind,lots of appearance,vast quantity and strong space-time relatively. It is a new aspect to develop using data mining method. Data mining is the process in which find information from database. The information is potential,unbeknown,valuable. Data mining is the central part of the KDD. It is introduced in chapter two. The keystone researched in this paper is to discover accident black spots from the database of traffic accident by the technology of data mining. The meaning of traffic black spots(prone location of traffic) are the spots or segment of road in which the number and the loss of the accident is higher than others. There is not a uniform agreement on the conception of traffic black spots all over the world. According the requirement analysis of the project, the conception of accident black spots defined in china has been put forward firstly in this paper. The spots in which excessive accident emerged is the place in which more than three fatal accident take place in the range of five hundred meters a year. And the segment is the place in which more than three fatal accident take place in the range of two thousand meters a year. Traffic black spots are introduced particularly in chapter three. On the basis of analysis of the main existing method, by means of comparing and analyzing, this paper put forward identification of prone location of traffic based in cluster analysis. Data mining used widely in ITS. Using cluster analysis, the cluster which has a similar characteristic can be found from database. It can find the area of large density by the Density-Based Algorithm. Traffic black spots are the place in which the density of accident is large, so it can be found out by clustering analysis. Prone location of traffic can be found out by the number of accident in a certain length road. The area can be considered as accident black spots in which the number of accident exceed the value confirmed by authority. That is the basic idea how to find out accident black spots with the number of accident. The method is used in our country. DBSCAN is a typical density-based algorithm for discovering clusters. The key idea is that for each point of a cluster the neighborhood(ε-neighbor) of a given radius has to contain at least a minimum number of points (MinPts). The advantage of DBSCAN is clustering at high speed, further more it can find the outliers and clusters in any form. Considering the characteristic of the algorithm, there are some similitude between the neighborhood and the length of road, the number of accident and the MinPts also. The algorithm can not only satisfy the requirement of government but also judge how badly the accident is by checking some information. The main advantage of it is to find the accident black spots in any place. By comparison, many algorithms can only check the traffic black spots in the place designed beforehand. The demonstration for the correctness of the algorithm is given in chapter four. There are three models of algorithm put forward in chapter four. The first model is based the DBSCAN. You should make sure the ε-neighbor and MinPts at first. Then search the cluster composed of the point whose neighborhood of a given radius has to contain at least a minimum number of points. This model of algorithm can satisfy the demand of government completely, but the loss of traffic accident is not considered. The second model add the step in which the result given by algorithm of the first model is examined for taking the losing into account. At first, each facet of accident must be given a weight for compute the losing. Then it looks over the cluster again and finds the cluster which has more average losing then the value given before. So this model of algorithm is composed of the first model and additive step. This model has a disadvantage that it can leak some traffic black spots because the density of accident is not uniform in every road. The average losing of these black spots is not high, but some place in which is high. The result can still satisfy the demand and provide advice to manager of traffic. So this model has extensive applicability. The algorithm improved in the third model to solve the problem in the second model. It also requires theε-neighbor and MinPts which is not important. For example it can be given the value two. The most important part in the algorithm is that the average losing of accident black spots is more than the value given before. At first the original cluster is found. Then searching the point of accident in whose ε-neighbor the average losing is satisfied. The loss of traffic accident is taken into account in the model. According to the three models, the module of find accident black spots is...