Some Advancement in Model Selection Methods And Its Application to a Genetic Epidemiology Study

選模方法之最近發展及其在遺傳流病研究之應用

連郡儀(Chun-Yi Lian)

指導教授 ： 熊昭 ； 謝文萍

國立清華大學/理學院/統計學研究所/碩士(2008年)

https://doi.org/10.6843/NTHU.2008.00591

摘要

Model selection is an important topic in data analysis. If the model is selected appropriately, we can use it to predict well. We introduced three tools for model selection in this paper. These three tools are QIC (Quasi-likelihood under the Independence model Criterion), L1-regularization path algorithm for generalized linear model, and L2-penalized logistic regression with a stepwise variable selection. The method QIC can be used for the correlated data such as family data. L1-regularization path algorithm and L2-penalized algorithm can be used for high-dimensional data such as microarray data. If we focus on gene interactions, the method L2-penalized algorithm may be useful. Our data from the SAPPHIRe (Stanford Asian Pacific Program for Hypertension and Insulin Resistance) project is from family data hence correlated. We use these three methods for the data set and compare the models selected by different methods and evaluate the performance of the prediction.

關鍵字

選模 ； HASH(0x1cd866f0) ； HASH(0x1cd86790) ； HASH(0x1cd86830)

並列摘要

HASH(0x1cd86b60)

並列關鍵字

model selction ； GEE ； QIC ； penalized likelihood

參考文獻

Efron, B., Hastie, T., Johnstone, I. and Tibshirani, R. (2004). Least angle regression. Ann. Statist., 32, 407-499.

Friedman, J. (1991). Multivariate adaptive regression splines. The Annals of Statistics 19, 1-67.

Golub, T., Slonim, D., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J., Coller, H., Loh, M., Downing, J., Caligiuri, M., Bloomfield, C. and Lander, E. (1999). Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science, 286, 531-537.

James C. and Guoqi Q. (2007). Selection of Working Correlation Structure and Best Model in GEE Analyses of Longitudinal data. Simulation and Computation, 36: 987-997.

Jianqing F. and Runze L. (2001). Variable Selection via Nonconcave Penalized Likelihood and its Oracle Properties. JASA, Vol. 96, No.456.