Regularization Paths for Conditional Logistic Regression: The clogitL1 Package.

Area of Science:

• Statistical modeling
• Machine learning
• Computational statistics

Background:

• Logistic regression is a fundamental statistical model for binary outcomes.
• Penalized regression methods like lasso and elastic net enhance variable selection and prevent overfitting.
• Conditional logistic regression is suitable for matched or clustered data.

Purpose of the Study:

• To implement and evaluate an efficient algorithm for fitting conditional logistic regression models with lasso and elastic net penalties.
• To compare the variable selection and prediction performance of conditional logistic regression against its unconditional counterpart.
• To demonstrate regularization path estimation and cross-validation for the conditional model.

Main Methods:

• Application of the cyclic coordinate descent algorithm.
• Incorporation of sequential strong rules for algorithmic speed-up.
• Simulation studies and real-world data analysis.

Main Results:

• The implemented algorithm offers significant speed improvements over standard methods.
• The conditional logistic regression model outperforms the unconditional model in variable selection on appropriately distributed data.
• Regularization paths and cross-validation strategies were successfully demonstrated for the conditional model.

Conclusions:

• The developed algorithm provides an efficient tool for penalized conditional logistic regression.
• Conditional logistic regression is a valuable alternative for variable selection when data exhibits conditional dependence.
• The methods presented facilitate robust model fitting and evaluation for complex datasets.