108/5/22(三)Professor Wei-Yin Loh (Department of Statistics, University of Wisconsin-Madison, U.S.A.)
Logistic Regression Trees
Professor Wei-Yin Loh
(Department of Statistics, University of Wisconsin-Madison, U.S.A.)
Logistic regression is a popular method for analysis of binary response data. It can be hard to beat in terms of prediction accuracy if the sample size and number of predictor variables are both small. But if the sample size or numbers of predictors is large, expert knowledge (such as decisions on transformations and interaction and high-order terms) is often needed for logistic regression to be efficient and effective. Worse yet, a logistic regression model containing many variables is usually hard to interpret. Finally, logistic regression cannot be applied to data with missing values unless they are imputed beforehand.
This talk introduces a new algorithm for building regression tree models where a simple linear logistic regression model is fitted to the data in each node of the tree. Because only one predictor variable is used in each logistic model, interpretation is straightforward without worry of difficulties caused by multicollinearity. Furthermore, because the algorithm can deal with missing values in the data, there is no need for their prior imputation.
日 期：108年5月22日(星期三) 16:10~17:00