建構以信令旅運曝光量與事故風險地圖為基礎之駕駛風險評估

Driver Risk Assessment based on Cellular Travel Exposure and Crash Likelihood Map

以駕駛行為基礎的車輛保險機制面臨諸多挑戰，如駕駛行為資料難以蒐集、駕駛行為與行車風險關聯待確認。基此，本研究發展以曝光量為基礎的車險機制，利用事故風險地圖與手機信令資料還原軌跡，藉此評估行車風險。本研究包含事故風險地圖之建構、地圖匹配與運具判別程序，以及評估用戶風險及與保費鏈結。事故風險地圖係利用事故頻次與嚴重程度建構預測模型，透過風險矩陣評估路段風險分數。地圖匹配與運具判別係還原駕駛人行車軌跡，並從中挑選使用私人運具者。最後，本研究疊合風險地圖與軌跡，計算60位調查用戶行經路段的總風險，換算為對應的保費調整係數。試算結果發現本研究研提之風險得點架構具鑑別力，未來可進一步驗證其與事故出險關聯。

The design of many Usage-based insurance schemes determines the insurance premium based on driving behaviors. The popularity of such schemes is hindered by many reasons. For example, the data of driving behaviors are hard to be obtained and the linkage between driving behavior and crash likelihood has not yet been validated. Therefore, this study proposes an exposure-based insurance framework based on crash likelihood map and user travel trajectories. The crash likelihood map is formed based on a jointly estimating crash frequency and severity model for each link of the map, and the travel trajectories are extracted from user cellular data by using map-matching algorithm and mode classification model. To develop and validate the travel trajectory extraction model, the cellular data and travel diary data of a total of 60 volunteers are then collected. Lastly, the relative risk scores of 60 volunteers are then calculated. The results show the calculated risk scores can discriminate the driving risk of these drivers, suggesting the applicability of the proposed framework. The relationship between risk scores and number of crashes deserves to be further investigated.