Summary: | 碩士 === 國立彰化師範大學 === 企業管理學系 === 98 === With the improvement of transportation means, the traffic problems become more complicated, and the accidents rate increases than before. The motorcycles with the features of economy and convenience have occupied nearly 70% of all registered vehicles. In addition, the fatal rate is ranked first among all kinds of vehicles. Therefore, how to reduce the accidents rate efficiently has become the most important issue. Under such circumstance, our study intends to apply data mining techniques to discover the hidden information.
Our study uses the fatal traffic accidents data during 2005 to 2007. First, the chosen independent input variables including age, speed, climate, light, and so on are summarized by the literature review, whereas major injure and degree of injury are the target variables by the classification and regression tree (CART) of SPSS 12.0 software. Second, eleven related variables are further taken into account as the input variables to discuss the impact of degree of injury. Later, use principal component analysis of SPSS 17.0 software to identify key variables, perform the same procedure based on key variables, and then compare the results in terms of accuracy. Finally, use the same variables for neural network of SPSS 12.0 software and then compare the results generated by neural network and CART. Finally, two target variables (major injury and degree of injury) are simultaneously put into neural network for analysis.
The results show that the accuracy of CART is higher than that of neural network. Moreover, there are different types of input variables resulting in different degree of injury but the key variables are not significantly different by CART or neural network. Finally, we also provide some suggestions for possibly improving the traffic environment.
Keywords: Traffic accident; Data mining; Fatal traffic accidents; Decision tree; Dimension reduction; Neural Network
|