A Classification Approach Based on Density Ratio Estimation with Subspace Projection

• Main Authors: ChingChuanChen, 陳慶全
• Other Authors: Ray-Bing Chen
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/77891250069168745439

碩士 === 國立成功大學 === 統計學系 === 102 === In this work, we consider a classification method based on density ratio estimation. Kanamori et al. (2009) proposed a direct estimation with least-squares approach for the density ratio estimation and showed how to use their density ration estimation approach for classification problem. However, the curse of the dimensionality would be caused the computational problem. To overcome this problem, we suggest projecting data into the proper subspace and then implement the density ratio estimation on this subspace instead of the whole data. We can choose to rotate data or basis. The latter is more efficient than the fronter. Simulation studies with different scenarios and several real examples are used to illustrate the performances of the proposed method. Based on the area under the receiver operating characteristic (ROC) curve (AUC) classification score, the results show the improvements of the proposed method and demonstrate the proposed method is comparable with other approaches, for example, logistic model approach. We also consider other classification score, partial AUC, the results presents that the proposed method performs fairly.