Application of LASSO regression in estimating B-Spline-Based hazard functions

• Main Authors: Lin, Zi-Yuan, 林子元
• Other Authors: Huang, Tzee-Ming
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/v47b98

碩士 === 國立政治大學 === 統計學系 === 105 === A strong assumption in the Cox proportional hazards model requires linearity of the covariates on the log hazard function. However, this assumption may be violated in practice. Alternatively, it is feasible to model the nonlinear effect via a combination of B-spline basis functions. In estimating the basis coefficients, the group lasso is applied. By so doing, a group of coefficients can be set zero simultaneously if the corresponding covariate is not predictive. Lastly, I develop hypothesis testing regarding this model. In addition to the ordinary Wald statistic, likelihood ratio statistic, and score statistic, two other types of testing statistic are considered: one adjust for penalty function and the other one based on bootstrap samples. Simulation studies are carried out to evaluate the performance of the proposed statistics.