連續性指標之分類與預測比較

Abstract

ROC 及PPV 曲線經常被用來評估連續指標的分類與預測性，在此論文裡面我們驗證一個指標擁有好的分類性會引申好的預測性，反之亦然。此外，我們根據不同的座標系及量測尺度提出一系列的檢定方法，更近一步藉由數值的模擬來評估所提出方法的執行力。最後我們將我們的方法應用在印地安婦女糖尿病研究及肝病研究上。

The receiver operating characteristic curves (ROC) and the positive predictive value (PPV) curve are often used to assess the performance of a continuous marker in classification and prediction, respectively. In this thesis, we showed that the better the marker in classification, the better the marker in prediction, and vice versa. Moreover, some test rules were established based on a variety of coordinate systems and metric measurements for the closeness between the ROC curves and that for the PPV curves. A class of simulation experiments were further implemented to investigate the performances of the developed inference procedures. In addition, our procedures are applied to two empirical examples from the studies of Pima-Indian diabetes and liver disorders.