以類神經網路建構彰化縣高齡者肇事責任與傷亡情形模式之研究

Abstract

由於過去研究結果顯示高齡者發生事故時與其他年齡層相比有較嚴重的傷亡程度及事故嚴重性，因此本研究選擇以高齡者作為事故特性的研究對象，並以101-103年彰化縣區車輛行車事故鑑定委員會事故案件資料作為本研究事故特性分析之基礎，透過本研究瞭解彰化縣高齡者事故發生特性，進而提出相關策略以期降低交通意外事故之發生。本研究所使用臺灣省彰化縣區車輛行車事故鑑定委員會事故案件共940筆，另新增近老年齡分類(60歲以上未滿65歲)，與初老(65-74歲)、中老(75-84歲)、老老(85歲以上)高齡者事故特性比較。排除當事者未滿60歲之案件共計609筆，事故當事人超過二人、重複鑑定或事故案件資料有缺漏者共計74筆，扣除掉不可用的683筆資料後可用於本研究事故特性分析資料共計257筆，利用卡方檢定中有9項變數(幹支道、事故位置、有無號誌、事故型態、兩車關係、高齡者是否超速、高齡者違規情形、第二當事人是否超速、第二當事人違規情形)與肇事責任有顯著影響關係；10項變數(光線、道路速限、高齡者性別、高齡者年齡、高齡者車種、高齡者是否超速、高齡者有無駕照、第二當事人性別、第二當事人車種、第二當事人有無駕照)與傷亡情形有顯著影響關係。依據上述卡方檢定顯著變數針對肇事責任及傷亡情形考量類神經網路方法建立肇事責任及傷亡情形之兩預測模式。其分析結果顯示，預測肇事責任模式之訓練及驗證判中率為81.1%及70.1%，前三項重要變數為高齡者違規情形、第二當事人違規情形及兩車關係；預測傷亡情形模式之訓練及驗證判中率為85.4%及80.3%，前三項重要變數為高齡者車種、高齡者有無超速及光線。透過研究方法得到之重要影響變數研擬彰化縣高齡者事故道安改善措施。

From the past studies show that the elderly has higher severity rate comparing to the other ages. Therefore, this study chooses the elderly as the study object, and selects the data from the Chang Hua Traffic Accident Authentication Committee (CHTAAC) from year 2012 to 2014 to analyze the characteristics of traffic accidents. It is desired that through this study finds out the characteristics of traffic accident in Chang Hua County and proposes the improving strategies to reduce the relevant traffic accident. There are 940 cases from this studying period, and the elderly are divided into four groups such as the near elderly (from 60 to 64 years old), the young elderly (from 65 to 74 years old), the middle elderly (from 75 to 84 years old), and the old elderly (above 85 years old). The useful data is 257 cases. Uses the Chi-square test to find out nine significant variables of the traffic accident liability are main/artery, locations, with/without traffic signaling, type of accident, relationship between two cars, spending of the elderly, traffic regulation violation of the elderly, speeding of the second party, and traffic violation of the second party. Ten significant variables of the traffic accident severity are lighting, speed limit, gender of the elderly, age of the elderly, type of vehicles of the elderly, speeding of the elderly, with/without driver license of the elderly, gender of the second party, type of vehicles of the second party, and with/without license of the second party. The nine significant variables were selected to construct the predicting model of the traffic accident liability by the Artifical Neutral Network (ANN) and ten significant variables were selected to construct the predicting model of traffic accident severity by ANN. The results show that in the model of traffic accident liability, the prediction accurate rate is 81.1 percent in training part, and the predicting accurate rate is 70.1 percent in the validation part. The top three significant variables are traffic regulation violation of the elderly, traffic regulation violation of the second party, and the relationship between two cars. Meanwhile, in the model of the traffic accident severity, the predicting accurate rate is 85.4 percent in the training part, and the predicting accurate rate is 80.3 percent in the validation part. The top three significant variables are type of vehicles of the elderly, speeding of the elderly, and lighting. Finally, this study proposed several improving strategies for Chang Hua County.