본 연구는 시내버스 운전자의 실제 운행기록 정보를 토대로 사고 잠재력이 높은 운전자를 사전에 판단할 수 있는 모형을 개발함으로써 향후 운전자 관리방안 마련을 위한 기초자료 제공을 목적으로 수행되었다. 운전자 운행기록자료에서 사고발생과 관련한 횡방향 가속도 등 6개의 가속도 유의변수를 도출하는 한편, 판별분석 및 로지스틱회귀분석을 적용하여 운전자 사고발생 예측모형을 개발, 모형간 정확도를 비교하였다. 분석결과, 판별분석에 의한 예측모형은 최대 62.8%의 비율로 사고발생 운전자 분류가 가능하며, 로지스틱회귀분석에 의한 예측모형은 최대 76.7%의 비율로 사고발생 운전자 분류가 가능한 것으로 나타났다.
According to the 2013 statistic data of the National Police Agency, the number of accidents by commercial vehicles accounts for 22.4% of total car accidents, and as compared to non-commercial vehicles, it shows a ratio more than four times higher than the number of registered vehicles. Of these, city bus that is an important means of transportation for local residents and students’ everyday lives and connects relatively short distances in the area takes up only 0.17% of the whole number of registered vehicles, but it was reported that more than 15 times more accidents occurred. Although the city bus accidents overall, show a downward trend, e.g. A decrease by about 50% from 2000 through 2013, but for continuous reduction in accidents, it is necessary to conduct an analysis of the causes of the accidents and provide a customized preventive measure corresponding to this, and especially, various studies are necessary for the management of drivers with a high accident potentiality. Thus, this study was conducted to provide basic information for preparing a plan for the management of drivers, by developing a model through which city bus drivers with a high accident potentiality could be judged in advance based on the information about their actual driving record. As a result of a discussion about the factors of car accident-related driver driving behavior and the characteristics of city bus accident, this study found out that the factors related to speed and acceleration are the main factors of dangerous driving behaviors and in-vehicle negligent accidents like overturn in operation account for a high proportion in city bus accidents. This study is an empirical study using the driver’s actual driving information as compared to the existing empirical and experimental studies of car accident-related driver factors, which drew significant variables related to accident occurrences, developed a model for the prediction of the occurrence of the accidents in the drivers and compared the accuracy of models, applying discriminant analysis and logistic regression analysis. As a result of a comparative analysis of the characteristics of driving behavior between a group with accident occurrence and a group without accident occurrence to draw the significant variables of accident occurrences, it was found that the difference between the groups was significant in six acceleration factors like lateral acceleration at a confidence level over 95%, and based on this, the six acceleration variables, including lateral acceleration were set up as significant variables for the model for the prediction of accident occurrences. While variables for analysis to develop a model through an analysis of the multi-collinearity between variables were selected, the accuracy of classification was analyzed, applying all variables except for those with suspected multi-collinearity to the development of a model. The statistical significance of significant variables in the model was secured, and through removing the variables with a low influence on the model sequentially, the variation of accuracy of classification of the drivers who might cause an accident. As a result of an analysis, the prediction model by the discriminant analysis could classify drivers who might cause an accident at a ratio up to 62.8%, while the prediction model by the logistic regression analysis could classify drivers who might cause an accident at a ratio up to 76.7%. As a result of the development of a model, variables of deceleration and acceleration acting in the right direction were drawn as the optimum variables of the classification of drivers according to accident occurrences, and as a result of a test of the validity of the model, the accuracy rate was 76.7%. As a result of a test of homogeneity by time, it turned out that the difference between the groups was not significant at 95% confidence level, so the validity of the developed model was secured. In addition, as a result of a test of the prediction ability of the model, it showed an accuracy rate of 84.1%, and as a result of a test of homogeneity by time in the predictive classified groups, it turned out that the difference between the groups was not significant at 95% confidence level. This study has significance that it presented a methodology by which the possibility of accident occurrences could be judged based on the information about the drivers’ actual driving, and in the future, it will be used as a good tool for judging the potentiality of accidents to be caused by city bus drivers in advance. Especially, it will be necessary to propose a strategic basis for the correction of the driver’s driving habits for the reduction of in-vehicle negligent accidents that account for a high proportion of city bus accidents and draw up a theoretical framework of the execution of safety training customized for each driver for the reduction of driving behaviors such as abrupt deceleration and hard right turn, while it is expected that it will be used as a base for the selection of drivers through an analysis of their driving record when new drivers are recruited in the future.
그림목차 ⅲ표목차 ⅳ국문초록 ⅷ제1장 서 론 1제1절 연구배경과 목적 1제2절 연구내용과 방법 3제2장 이론적 고찰 5제1절 시내버스 교통사고 현황 51. 우리나라 교통사고 발생추이 52. 사업용자동차 교통사고 발생현황 72.1 업종별 교통사고 발생현황 102.2 업종별 교통사고 사망자수 113. 시내버스 교통사고 현황 및 특징 12제2절 교통사고와 운전자 변인 151. 교통사고 관련 인적요인 152. 운전자 위험운전행동 요인 17제3절 시사점 및 연구의 방향 221. 기존 연구의 한계점 222. 연구의 방향 23제3장 분석용 자료구축 및 연구 방법론 설정 24제1절 분석용 자료구축 241. 디지털운행기록 자료의 구조 및 형태 242. 분석용 자료 구축 273. 분석용 자료의 일반적 특성 28제2절 분석방법론 검토 311. 개요 312. 분석방법론 322.1 판별분석(Discriminant Analysis) 322.2 로지스틱회귀분석(Logistic Regression Analysis) 343. 분석절차 및 방법 37제4장 모형개발 및 검증 39제1절 유의변수 선정 391. 분석용 변수의 구성 392. 유의변수 선정 423. 다중공선성 64제2절 사고예측 모형개발 및 검증 661. 모형개발 661.1 모형개발 방법론 661.2 모형개발 과정 682. 모형검증 94제5장 결 론 96제1절 결론 96제2절 향후 연구과제 99참고문헌 100Abstract 104
