Stability selection for lasso, ridge and elastic net implemented with AFT models

Area of Science:

• Biostatistics
• Statistical Modeling
• Machine Learning

Background:

• Instability in model selection is a significant challenge with high-dimensional datasets.
• Variable selection methods often struggle with large numbers of covariates and censored data.
• Accelerated failure time (AFT) models are effective for data with heavy censoring and high dimensionality.

Purpose of the Study:

• To evaluate the efficacy of stability selection in improving variable selection performance for censored data.
• To compare the performance of Lasso, ridge regression, and elastic net with and without stability selection.
• To assess the impact of data dimensionality and covariate collinearity on selection stability and performance.

Main Methods:

• Implementation of stability selection using Lasso, ridge regression, and elastic net.
• Application of these techniques to censored data within the framework of AFT models.
• Performance evaluation through simulation studies and analysis of two real-world datasets (breast cancer and diffuse large B-cell lymphoma).

Main Results:

• Stability selection consistently provides stable variable selection outcomes.
• Methods incorporating stability selection demonstrate improved performance compared to those without, particularly as data dimensionality increases.
• The benefits of stability selection are observed irrespective of collinearity among covariates.

Conclusions:

• Stability selection is a robust technique for enhancing variable selection in high-dimensional, censored regression models.
• The integration of stability selection with regularized regression methods (Lasso, ridge, elastic net) offers superior performance and stability.
• This approach is particularly valuable for complex biological and medical datasets where high dimensionality and censoring are common.