A Traffic Accidents Analysis Model Based On Data Mining Theory

In recent years, with the enhanced speed of vehicles and increased traffic volume, road traffic accidents (especially fatal accidents) show a increasing gradually tendency, that has arrested some relative department's attention gradually. As researches currently show, in most domestic road traffic accident analysis systems, the decision-making remains in manual handles, which is a primary reason of inefficient decision-making from huge amount of data. So it is necessary to make scientific researches and effective improvement to road traffic accident analyses. As a traffic management department, traffic police are engaged on data analyses of accidents, and manage to discover something useful and have a definite object in view. In the face of huge amount of data of traffic accidents in the past years, it's imperative to find a kind of data mining technique. In recent years, data mining has risen a great attention of the information industry, and the main reason is because data ocean become more and more huge we need new technology to transmute the very large data into useful information or knowledge. Rough Set theory is founded by a Polish mathematician—Z.Pawlak in 1982, which is a new kind of mathematical tool to deal with Vagueness and Uncertainty problems. The main advantage of Rough Set theory is that it needs nothing about any preliminary or additional information of data. Effective reduct algorithms are foundations to apply rough set theory to data mining and knowledge discovery in database, but it has been proved that the problem to obtain all reducts or the minimal reduct is NP-hard problem. Interiorly, it is seldom to apply rough set theory to actual fields, especially in traffic field. In chapter 1 and chapter 2, we make a summary of data mining and rough set theory, then combine the idea of association rule and propound a kind of decision rule reduct algorithm based on preferential information. We apply the algorithm to actual project to analyze causes of traffic accidents and finish the implement of Traffic Accidents Analysis System. In chapter 3, we introduce designs for the project and functional models. The main aim of the project is to identify road black spots and find causes of accidents, by analysing historical data of traffic accident database of ChangChun City, and assist traffic department decision-makers to make decision. The system is composed of four models, including Accident Position Record,Identification of Black Spots,Data Statistic and Decision Analysis(shown in Figure 3.1). The first model provides data for other three models. Combining GIS(Geographic Information Systems), it makes a standardization for accident data via electronic map and transforms accident place of initial data to position identifiable by electronic map. The second model is to analyse accident instances of all areas or roads and to identify black spots via the position information of data. The third model is to observe relations between accident causes or accident shapes and road conditions or environment by means of statistics. The last model applies the forenamed decision rule reduct algorithm to the accident data and obtains decision rules, then finds out the impact of road conditions and environment on accident causes and accident shapes. In chapter 4, we quote a concept —"preferential information"and propound the decision rule reduct algorithm based on preferential information. The main idea of the algorithm is to use iterative strategy and apply some cycle operations to a initial character set, such as judging, intersecting, rule extracting etc, finally obtain decision rules with support and confidence according with users. First we describe decision tables in the form of association rule and describe equal classes of decision tables by citing character and concept. Then we add a frequence attribute into decision tables to count each object, which belongs to neither condition attributes nor decision attributes. We can establish an one-meta character set via each condition attribute value and define delete operation and intersect operation of character sets. Finally we can obtain decision rules set by means of iterative delete operation and intersect operation on initial character sets. By setting minimal support threshold αand minimal confidence threshold β, we can obtain decision rules according with users. The setting of value of minimal support and minimal confidence is crucial to the number of final obtained rules. Too much rules may cause too low efficiency and make too much insignificant rules, but too few rules may lose some useful...