以二元迴歸方法建構建物震害危險度預測模式之研究－－以中興新村都市計畫地區為例

ESTABLISHINGGTHEGFORECASTGMODELGOFGTHE HAZARDGOFGTHEGBUILDINGGINGTHE EARTHQUAKEGDISASTERGBYGBINARYG REGRESSIONGMETHOD─CASEGSTUDYGATGTHE URBANGPLAANINGGREGIONGOFGCHUNG-HSING NEWGVILLAGE

台灣於921 大地震後，民眾對於地震有很大的恐懼，而在地震災害特性中，人員的傷亡大多因為受建物毀損及倒塌而造成，因此，建物之抗震能力成為防震之最基本要件。影響建物抗震能力的因素可分為兩大類：自然環境特性及建物本身特性，而本研究係以二元迴歸方法建構建物危險度預測模式，並透過係數來判定建物毀損與建物特性之關係。本研究以中興新村都市計畫地區為研究範圍，透過相關文獻及有關單位之921 建物毀損調查資料，歸納出建物震害原因；而所選取模式之變數，則考量資料取得之可行性及實地之調查，再以二元迴歸方法之羅吉斯迴歸（Logistic Regression）來建構建物震害危險度模式，經由模式變數之係數與預期符號之對照、係數顯著性檢定、預測準確率來了解模式中各參數所顯示出的意義及其在政策上的應用。

After 921 earthquake in Taiwan, people have a great fear of earthquakes. In terms of disaster characteristics in earthquakes, casualties were caused by breakages and collapses of the buildings. The ability of the building to resist earthquakes becomes a fundamental requirement for earthquakes prevention. The factors which influence the ability of a building to resist earthquakes can be categorized into two: the characters of natural environment and the characters of the building itself. Binary Regression Method was used to construct the forecast model about the hazard of the building in the earthquake disaster, and to differentiate the relationship between the damages of the building and the characters of the building by the coefficients in the model. The Urban Planning Region of Chung-Hsing New Village is the researched subject. The reasons for the building damages in the earthquake disasters can be induced by relative researches and investigation data of the buildings damages in fields collected by many institutions during 921 earthquake time. The model variables are chosen by considering the feasibility of the data which are obtained and investigated in fields. A forecast model about the hazard of the building in earthquake disasters is established by Logistic Regression of Binary Regression Method. The significance of the parameters of the model and the application to the strategy are found out through the contrast between coefficients of model variables and the expected sign, the test of significance of the coefficients and accurate prediction percentage.