Time-varying feature selection for longitudinal analysis

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Traditional model selection often uses global features, which may miss time-dependent covariate effects.
• Understanding time-dependent relationships is crucial in many scientific fields, including health studies.

Purpose of the Study:

• To propose a novel time-varying coefficient model selection and estimation method using splines.
• To develop a penalty function that utilizes local-region information for improved estimation.
• To demonstrate the method's utility in capturing time-dependent covariate effects.

Main Methods:

• Utilizing a spline-based approach for time-varying coefficient models.
• Introducing a new penalty function that incorporates local-region information.
• Conducting simulation studies to compare with global feature selection methods.
• Applying the method to a longitudinal growth and health study.

Main Results:

• The proposed model selection method incorporating local features demonstrates superior performance compared to global feature selection approaches.
• The spline-based method effectively captures time-dependent covariate effects.
• The method is validated through simulation studies and a real-world health dataset.

Conclusions:

• The proposed local feature-based spline approach offers a powerful tool for time-varying coefficient model selection and estimation.
• This method is particularly advantageous when scientific interest lies in detecting localized, time-dependent covariate effects.
• The approach provides a more nuanced understanding of dynamic relationships in longitudinal studies.