Classification of longitudinal data through a semiparametric mixed-effects model based on lasso-type estimators

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Classical Bayes classifiers require well-modeled densities.
• Longitudinal data analysis presents unique modeling challenges.
• Predicting pregnancy outcomes from early hormone levels is critical.

Purpose of the Study:

• To propose a novel classification method for longitudinal data.
• To develop an efficient estimation procedure for semiparametric models.
• To predict normal versus abnormal pregnancy outcomes using hormone data.

Main Methods:

• Developed a semiparametric linear mixed-effects model (SLMM).
• Proposed a unified estimation procedure using a penalized EM-type algorithm.
• Employed a lasso-type procedure for nonparametric estimation of the unknown function.

Main Results:

• The proposed SLMM offers an efficient estimation method.
• The method was validated through simulation studies.
• The model successfully predicted pregnancy outcomes on real-world hormone data.

Conclusions:

• The penalized EM-type algorithm provides an efficient estimation for SLMM.
• The proposed method is effective for classifying longitudinal data.
• This classification approach has practical applications in predicting pregnancy outcomes.