| Summary |
Je-wei Yang, 2006.08
Feng Chia University - Graduate Institute of Traffic and Transportation Engineering and Management
Most traffic accident analyses and authentication results consider human errors as the main factor, or even the only factor, which induced the occurrence of traffic accidents. However, in fact, the approaching roadside environment that drivers confront right before crash can sometimes play more important role than human error on the occurrence of the crash. Inappropriate geometry design, poor pavement skid resistance, and inadequate sign and signal arrangement can all create a dangerous driving environment. Consequently, putting all the blames to drivers is unfair and similar accidents may occur one after one. Apparently, there is a need to study roadside environment. On the other hand, due to the fact that great amount of traffic accidents occurred at intersections, it is thus worthwhile to study the correlation between nonhuman factors around intersections and traffic accident occurrence. Such information may help agencies to figure out what needs to be done and what should be done first. It may also serve as a reference for future intersection design and management.
Factors related to traffic accidents are generally inter-correlated, multivariate regression was thus adopted to analyze traffic accident problems.Yang used too many factors in that model, which made that model difficult to use. In this study, 18 out of the 53 factors used in Yang’s model were first sieved out by using statistical dependence test, chose one hundred and two intersections in Taichung City, divided them into one thousand and thirty-nine moving direction combinations, and constructed the forecast model of the number of the intersection traffic accidents with the data of 3,441 traffic casualties happening within four years.
Therefore, the study adopted the methods of the Neural Networkss and the negative binomial regression. The outcome was that the Neural Networks had the better forecast ability. The correlation coefficient was 0.994; the misjudging rate was 13.53%; the maximum negligence was 1.37; the networks' MSE value was 2.98 ×10-6. Prediction results of these models were not all very accurate, yet can be considered satisfactory. Finally, 10 out of the 18 factors were sieved out using sensitivity analysis and elasticity analysis. The study used the method of the classification and regression trees to determine intersections dangerous judgement. The outcome was that the Gini tree better than Twoing tree. Being found by the importance variable, the geometry was more important than traffic signs and traffic movement. The order were total road width, moving flow, motorcycle rate, turn left/right vehicle rate, total intersection volume of traffic, lane partition, speed limit or warning sign and cycle. |
