樣條函數的靈活性在於節點位置和數量作為自由變數，使得其在配適過程中具有多樣的功能。本研究深入探討樣條邏輯斯迴歸(spline logistic regression)中的兩種邏輯斯函數，致力於建立在不同節點配置下的對比效果。為了提升模型選擇的效率，我們對傳統的AIC向後選取法(backward AIC)進行了改良，特別適用於樣條邏輯斯迴歸。

我們的改良方法不僅能夠同時針對不同邏輯斯函數下的節點進行篩選，同時透過優化參數和變數篩選方法，相較於傳統AIC向後選取法，獲得更為優越的效果。此外，本研究進一步模擬了不同初始節點下的篩選效果，並使用ISE(integrated squared error)的平均值來評估各種情況下的模型配適效果。這項模擬的結果有助於更全面地理解不同初始節點配置對模型性能的影響，並提供對於節點選擇的方向。整體而言，本研究綜合運用不同方法，致力於提升樣條邏輯斯迴歸的節點選擇和估計效能。

The flexibility of spline functions lies in the freedom to choose the positions and number of knots, which endows them with diverse capabilities during the fitting process. This study delves into two different logistic functions within spline logistic regression, aiming to establish a comparative analysis under various knot configurations. To enhance the efficiency of model selection, we have improved the traditional backward AIC method, making it particularly suitable for spline logistic regression.

Our improved method not only allows for the simultaneous selection of knots under different logistic functions but also achieves superior performance compared to the traditional backward AIC method through parameter optimization and variable selection techniques. Additionally, this study simulates the selection effect under different initial knot configurations and uses the mean integrated squared error (ISE) to evaluate the model fitting performance in various scenarios. The simulation results contribute to a more comprehensive understanding of how different initial knot configurations affect model performance and provide guidance on knot selection. Overall, this research integrates various approaches to enhance knot selection and estimation efficiency in spline logistic regression.

S. Arifin, A. Islamiyati and E. T. Herdiani. Ability of ordinal spline logistic regression model in the classification of nutritional status data, Commun. Math. Biol. Neurosci., Article ID 83, 2023.

A. Yonar and H. Yonar. An Efficient Hybrid Algorithm with Particle Swarm Optimization and Nelder-Mead Algorithm for Parameter Estimation of Nonlinear Regression Modeling, Gazi University Journal of Science, vol. 35, no. 2, pp. 716–729, 2022.

J. A. Nelder and R. Mead. A Simplex Method for Function Minimization, The Computer Journal, vol. 7, Issue 4, pp. 308–313, January 1965.

F. Bessaoud, J.-P. Daures and N. Molinari. Free knot splines for logistic models and threshold selection, Computer Methods and Programs in Biomedicine, vol. 77, no. 1, pp. 1-9, January 2005.

E. J. Malloy, D. Spiegelman and E. A. Eisen. Comparing measures of model selection for penalized splines in Cox models, Computational Statistics \& Data Analysis, vol. 53, no. 7, pp. 2605-2616, May 15, 2009.