| Summary: | 碩士 === 國立政治大學 === 統計學系 === 106 === Spline functions are often used to approximate smooth functions. In nonparametric regression, if we use a spline function to approximate the regression function, selecting appropriate knots for the spline function will yield better fitting results. In this study, I consider three methods for knot selection: elastic net, LASSO and the UNIF method in [5]. Simulation experiments have been carried out to compare the performance of the three methods. From the simulation results, we have found that when the true regression function is smooth, knot selection base on elastic net gives better results. When the true regression function has large variation, knot selection base on the UNIF method gives better results.
|
|---|
