2909. The Hyperbolic Lasso: a regularized regression model with applications to health data
Invited abstract in session MC-43: Digital Philosophy, Academic career models and OR, stream OR and Ethics.
Monday, 12:30-14:00Room: Newlyn GR.07
Authors (first author is the speaker)
| 1. | Vinicius Layter Xavier
|
| Instituto de Medicina Social Hesio Cordeiro (IMS); Programa de Pós-Graduação em Ciências Computacionais e Modelagem Matemática (PPG-CompMat), State Universit. of Rio de Janeiro | |
| 2. | Claudia Jakelline Barbosa Silva
|
| Programa de Pós-Graduação em Ciências Computacionais e Modelagem Matemática (PPG-CompMat), Universidade do Estado do Rio de Janeiro, Instituto de Matemática e Estatistica |
Abstract
The Lasso regression method has been widely studied due to its characteristic of selecting a subset of variables, thus avoiding the problem of overfitting. The constraint that defines the Lasso is a non-differentiable optimization problem. We propose a differentiable formulation based on hyperbolic smoothing and the Hyperbolic Penalty method, thus allowing the use of continuous optimization methods. The lasso constraint is defined by the sum of the norm of the coefficients being less than equal to a constant t. Usually, the problem is solved indirectly, for different values of the lambda parameter. The proposed formulation has the characteristic of solving the constraint for specific values of t, thus allowing a better search for the possible values that the constraint can assume. The method's efficiency is illustrated by the use of healthcare datasets.
Keywords
- Machine Learning
- Computational Biology, Bioinformatics and Medicine
- Programming, Nonlinear
Status: accepted
Back to the list of papers