| 摘要: | 交通事故發生乃由多方構成，不同情境、時間、區域下所發生的交通事故種類嚴重性皆有所不同，只有正確適合國內交通環境的肇事評估分析與分類方法配合一套正確完備的交通事故紀錄檔案與易肇事目標之評定制度外，才能使交通安全順利且有效的進行。 本研究將資料探勘應用在此交通安全領域，期望指出，潛伏於肉眼下的肇事的影響因素。藉由分析肇事資料之人為、車輛機械、天候環境、道路設計間的關聯性，利用資料探勘方式之集群化技巧，挖掘隱藏於大量肇事資料間的因子關係與相對嚴重值。利用集群化找出同質性最高的肇事環境，以此結果為基礎分析，並以肇事危險度為分級標準與預測目標，結合多變量分析之判別分析驗證集群之正確性，以期建立一適合台灣地區之肇事危險模式。 在實證研究上，以民國九十二年間發生於台灣地區各公路系統肇事事件共114177件為研究對象。研究結果發現，在肇事者維度方面上肇事危險程度主要由駕駛資格、保護裝備與年齡情形不同所影響。在環境維度方面上肇事危險程度主要由號誌動作、路面狀態與光線情形不同所影響。在道路維度方面上肇事危險程度主要由分向設施的不同所主導，最危險等級則由快慢車道間的分隔情形所影響。在車輛維度方面上車輛用途情形隨著危險度增加降低影響度，車種的不同則隨著危險度增加而提高影響程度。 在肇事危險判別模式方面，肇事危險分為7層級時，其效率最大正判率最高，隨後便隨著分級數的增加而遞減。此模式正判率達79.6﹪，可知此模式之有效度，證明四維度之維度考量的確可有效衡量肇事危險。肇事危險程度影響比重依序為車輛維度危險、道路維度危險、環境維度危險、肇事人維度危險。而各維度危險度的增加也同時增加肇事危險程度，尤其是車輛與道路維度更為明顯。 Traffic accidents happen in many ways. The severity of traffic accidents vary by different situations. Suitable traffic analyzing and categorizing methods depends on correct and complete traffic accidents files. In this research, data mining is applied to the traffic safety field and expects to point out the hidden factors of traffic accidents. By analyzing the relationship among people, vehicle, weather, and the design of roads, we utilize the skills of clustering to find out the relative severity and the relationships between factors of traffic accident. Regarding Accident Injury Severity as hierarchical standard and predictive goal and use Discriminant Analysis of multivariable analysis in the hope of setting up a suitable Severity Categorizing Model for Taiwan. In the past research, 114177 traffic accidents happen on every highway system in Taiwan in 2003. The result of the study finds, in aspect of human factors, the accident injury severity is determined by the drivers’ qualification, equipments, and ages; in aspect of environment factors, the accident injury severity is determined by the movements of the signal, the state of road surface, and lights; in aspect of road factors, the highest class of danger is influenced by the separating among the speed lanes; in aspect of vehicle finds the effect of vehicle usage decreases with increasing dangerous degree, but different kinds of vehicles increases with increasing dangerous degree. In Accident Injury Severity Discriminating and Categorizing Model, 7 levels of dangerous show the greatest declaring rate of its efficiency, and then decrease progressively with levels increased. Because the sentencing rate is up to 79.6%, we can know how effective this way is and prove that the four-aspect Severity Categorizing Model can really weigh Accident Injury Severity. In Severity Categorizing Model, the influence percentage of four-aspects from high to low is aspect of vehicle danger, aspect of road danger, aspect of environmental danger, and aspect of people danger. And the raise of each enhances the Accident Injury Severity at the same time, especially the aspect of vehicle and road. |
