| Summary: | 碩士 === 逢甲大學 === 統計學系統計與精算碩士班 === 106 === Outlier detection is one of the most important steps in quality control. Many statistical algorithms have been developed for detecting outliers, but not all methods can be applied in general cases. In this study, we focus on the methods to deal with outliers in linear regression model. Some traditional indexes, including standardize residuals, externally standard residuals, predict interval, DFFITS, Cook’s distance are compared with the Bayesian predicted posterior probability approach. Moreover, we extend this issue to consider the information from multiple explanatory variables. Bayesian model averaging method with Markov chain Monte Carlo model composition is adopted. All algorithms are evaluated via simulations and temperature data from Central Weather Bureau.
|
|---|
