以類神經網路探討都市地區肇事嚴重程度

Abstract

提升道路交通安全所需考慮的層面相當廣，涉及工程、教育、執法等數個政府部門，礙於政府經費有限，如何以有限的資源完成最大的效益，將是首先面臨的問題。依據以往經驗，大部分交通事故主因多可歸類為「人」的因素，故對警察機關而言，執法作為雖可治標的、速效的降低肇事，然而執法強度過高卻導致民眾抱怨，且所投入過多的警力亦有浪費之虞，如可找出會導致嚴重傷亡交通事故之違規類型，並針對該類違規加強執法，應可有效達成防制肇事、減少警力浪費及提高民眾滿意度之三項目標。 隨著人工智慧的興起，各類運用人工智慧的方法去探求交通事故件數、嚴重程度及其影響因素間相互關係之研究也因應而生，其中類神經網路被運用的次數及準確度不下於常用的統計方法或人工智慧方法。因此，本研究採用該方法，構建一個類神經網路模式，用以預測出某路口、路段在發生肇事後，當事人之傷亡程度，測試結果顯示，路口肇事嚴重程度預測模式對於一般受傷案件的預測正確率為95%，重傷、死亡案件的預測正確率為42%，而路段肇事嚴重程度預測模式對於一般受傷案件的預測正確率為92%，重傷、死亡案件的預測正確率為45%，顯見類神經網路方法所構建的預測肇事嚴重程度模式有相當良好的績效。另外將模式實際應用於易肇事道路，所預判出會造成嚴重傷亡肇事案件的違規類型，可及早對管理者提出警訊，並研擬適當的行政作為加以因應。

To promote traffic safety needs to consider many aspects, including several government departments such as engineers, education and law enforcement. The first problem is how to use the limited resource and create the maximum efficiency. According to the experience, most traffic accident cases are caused by human factors. To police department, law enforcement can temporarily resolve the problems and quickly decrease the accident cases, but it not only leads to complain by people but also waste the police human resources. We can effectually prevent the accidents, low the waste of police human resources and raise the satisfaction of people by finding out and preventing the types of violations causing people to be killed or seriously injuredin road accidents. Nowadays, AI is used for a number of different reasons like searching the traffic accidents, analyzing the degree of severity, and researching the affection between cause and effect. The research builds a model by Neural Networks to predict the severity of injury resulting from traffic accidents at some intersection or on some section of the road. Experiment results reveal that at the intersection the general injury cases is 95﹪and the serious injury and death cases is 42﹪；on section of the road the general injury cases is 92﹪and the serious injury and death cases is 45﹪.Apparently, predicting the severity of injury resulting from traffic accidents model built by Neural Networks works well. The model applied to the road easily happening accidents can predict the types of violations causing serious injury and death cases. We can offer the warning to the administrator in time and devise proper administration.