SeminarsModel Selection in Regression Analysis: Classical Development
141
reads
Due to rapid development in large dimensional data acquisitions such as microarray, scientists look for useful methods on predicting a quantitative measurement when number of predictors p is much greater than n, the sample size. In this talk, we will present classical approaches in model selection through estimated prediction error for linear regression analysis when n is much larger than p. In particular, we discuss Mallows’ Cp (AIC), BIC, and cross-validation.
In addition, we address the selection of penalty for ridge regression which no longer gives unbiased estimators of regression coefficients. This leads us to next week’s discussion on model selection with p is greater than n.
reads
Hung Chen
2008-09-19
15:30:00 - 17:00:00
405 , Mathematics Research Center Building (ori. New Math. Bldg.)
Due to rapid development in large dimensional data acquisitions such as microarray, scientists look for useful methods on predicting a quantitative measurement when number of predictors p is much greater than n, the sample size. In this talk, we will present classical approaches in model selection through estimated prediction error for linear regression analysis when n is much larger than p. In particular, we discuss Mallows’ Cp (AIC), BIC, and cross-validation.
In addition, we address the selection of penalty for ridge regression which no longer gives unbiased estimators of regression coefficients. This leads us to next week’s discussion on model selection with p is greater than n.