Abstract:

In this study, we present a revised version of the Alternating Direction Method of Multipliers (ADMM) algorithm, specifically developed to enhance the convergence and stability of logistic regression models. Our updated technique achieves substantial enhancements in classification accuracy and model interpretability by integrating supplementary regularization and stability components into the ADMM update equations. The experimental findings on synthetic datasets demonstrate that our improved ADMM algorithm surpasses the conventional ADMM, offering a resilient and efficient solution for logistic regression. In this paper, we emphasize the potential of our adjustments to be a valuable improvement in optimizing logistic regression algorithms.