飛航安全為航空業者首要重視的議題,而提升飛航安全最具體的作為便是降低未來飛安事件發生次數,達到「事前預防勝於事後補救」之效。本研究應用多元複迴歸分析、倒傳遞類神經網路、支援向量機等資料探勘技術,在不同事件主題下,預測未來可能之各事件主題頻率值,以期達到預警之效,更進ㄧ歩建立飛安維護之循環體系。本研究主要分為兩階段,第一階段在於找出與各事件主題相關之指標,再將這些相關指標於第二階段裡建置各事件主題的預測模式。由於本研究採用三種資料探勘預測技術,因此將進一步比較不同技術之表現優劣,並以統計方法評估預測技術的適配性,證實不同分析技術的預測能力確實具有差異。在預測資料的建構上,採用資料群組化的方式,以提高模式之解釋能力,本文亦證實採用資料群組化的預測模式確實有較好的解釋能力,準確率亦明顯提升。 透過方法的比較,我們認為在不同的飛安事件風險預測中,倒傳遞類神經網路及支援向量機二種技術將能產生較適當的飛安預測模式,本研究重視整體的飛安考量,強調的是航空業者的飛安控管能力,因此,除模式的預測能力外,模式應用層面與後續的查核維護,亦將加以說明。
This research details the development of civil aviation risk analyze models. We define the aviation risk as the frequency of aviation events. Based on the FSMIS (Flight Standard Management Information System) database developed by CAA (Civil Aviation Administration), three prediction data mining techniques, multiple regression analysis, back propagation neural network, and support vector machine, are used to create prediction models for different type of aviation events. The FSMIS database contains the information of air operators’ safety performance in various inspection items. Using the information, the relationship between events and operation performance is analyzed and the event prediction models are created. According to our study results, it appears that a data smoothing strategy is very effective at enhancing the predictive accuracy of the models. The methodological comparison suggests that back propagation neural network and support vector machine offer a more promising technology in prediction aviation risk.