Dimension Reduction for Bayesian Classification Analysis

• Main Authors: Lee, Yun-Ju, 李韻如
• Other Authors: Lue, Heng-Hui
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/7m64np

碩士 === 東海大學 === 統計學系 === 105 === In literature, there are many existing methods for classification analysis. In this study, we use Bayes classifier for analysis. Due to the problem with high dimensionality, the classification analysis has become more difficult. To overcome such problem, we use dimensional reduction to relieve the difficulty we encounter. We propose to project data onto the effective dimension reduction space via IRT-PHD (Lue, 2015) and find significant variables based on t-ratio (Lue, 2004) and SIRI (Jiang and Liu, 2014). We classify data using Bayes classifier and demonstrate the advantage of our proposed method compared to SIR, SAVE and PHD through simulation study and analysis of real data set.