| Summary: | 碩士 === 逢甲大學 === 運輸科技與管理學系 === 99 === Recent years, traffic accidents have been increasing constantly in Taiwan, and the accident will cause the casualties, legal disputations, vehicle damages and other issues. However, domestic Local Traffic Accident Appraisal Committees (LTAAC) is lack of manpower, and each committee member has different background. Therefore, the different LTAAC have different judging criteria for traffic accident liability. It highlights that we need a rational criteria for those different LTAAC for reference. The aim of this study used the Classification and Regression Tree (CART), Genetic Programming (GP) and Artificial Neural Network (ANN) to construct an expert support system of traffic accident liability for pedestrian. Methodology includes literature review to select variables and then to apply the chi-square test to select the significant variables such as LTAAC database has 30 significant variables, and Taiwan Provincial Traffic Accident Appraisal Committee (TPTAAC) database has 25 significant variables, and both LTAAC and TPTAAC having the same traffic accident liability has 24 significant variables. We apply those significant variables to construct three models and then to validate.
There are 685 pedestrian accidents in the database, and the collision types include car/pedestrian, and motorcycle /pedestrian. Also, the database will be distinguished into three types: (1) for LTAAC data only; (2) for TPTAAC only; (3) the same traffic accident liability data between LTAAC with TPTAAC. This study shows that ANN has the maximum accuracy rate 87.25%. The next, CART has 82.35%, and the GP has 69.61%. This study offers an expert support system, so the relevant person from LTAAC or TPTAAC can input the significant variables from a specific traffic accident. Then the system can automatically output the traffic accident liability. It is hopeful that this expert system can be used to improve the traffic accident liability among LTAAC or TPTAAC.
|
|---|
