More efficient approximation of smoothing splines via space-filling basis selection

Area of Science:

• Statistics
• Computational Statistics
• Nonparametric Regression

Background:

• Smoothing spline estimators are powerful tools in nonparametric regression.
• High computational cost of full-sample smoothing splines limits their practical application.
• Existing approximation methods using random basis selection have limitations.

Purpose of the Study:

• To develop a more efficient basis selection method for smoothing spline estimators.
• To reduce the computational complexity of smoothing splines.
• To improve the approximation accuracy and prediction performance.

Main Methods:

• Proposed a novel basis selection strategy by choosing basis functions corresponding to approximately equally spaced observations.
• Conducted asymptotic analysis to evaluate the theoretical properties of the new estimator.
• Applied the method to synthetic and real-world datasets for empirical validation.

Main Results:

• The proposed method significantly decreases the essential number of basis functions required.
• Asymptotic analysis shows a reduction in computational complexity.
• Empirical results demonstrate smaller prediction errors compared to existing methods.

Conclusions:

• The proposed basis selection method offers a computationally efficient and accurate approximation for smoothing splines.
• This approach enhances the broad applicability of smoothing splines in large-scale data analysis.
• The method provides a practical solution for reducing computational burden without sacrificing accuracy.