Accident type-based analysis of risk and collision factors using data mining techniques
Date Issued
2015
Date
2015
Author(s)
Hsiao, Wei-Lun
Abstract
The conclusion of the past traffic safety study revealed that safety diagnosis and collision factors study on the basis of intersection faced the problem of detail lost for each collision type. It is essential to analysis traffic safety based on collision types and approaches. In addition, safety analysis using collision factors is reactive, the shortcomings include the less damaging collision may not be reported, small data quantity and the analysts need to wait for accidents to take place in order to prevent. Conflict events are more frequent than collisions and it brings complementary information. With surrogate safety measures, the road safety analysis can be proactive. This thesis presents a collision type based study on approaches using conflict and collision data. K-means algorithm is used to identify risk groups of approaches with similar conflict and collision attribute, decision tree and random forest are used to analyze the relationship between road attribute and risks outcome. The result shows the effectiveness of post encroachment time as a surrogate measure in right angle collision and right turn with through collision, with the best results at a threshold as 3s. In road design, the decision tree and random forest confirms the importance of collision factors: intersection angle, lane design, road markings and signal timing design. The thesis also established risk thresholds for post encroachment time and conflict rate as a diagnosis tool to evaluate traffic safety of approaches.
Subjects
Traffic safety
Surrogate safety analysis
Post encroachment time
Data mining
Collision factors
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-104-R02521508-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):12bbd7d8182149442b4e97aa905757e9