References & Citations
Statistics > Methodology
Title: Identifying a minimal class of models for high-dimensional data
(Submitted on 2 Apr 2015 (v1), last revised 25 Nov 2015 (this version, v3))
Abstract: Model selection consistency in the high-dimensional regression setting can be achieved only if strong assumptions are fulfilled. We therefore suggest to pursue a different goal, which we call a minimal class of models. The minimal class of models includes models that are similar in their prediction accuracy but not necessarily in their elements. We suggest a random search algorithm to reveal candidate models. The algorithm implements simulated annealing while using a score for each predictor that we suggest to derive using a combination of the Lasso and the Elastic Net. The utility of using a minimal class of models is demonstrated in the analysis of two datasets.
Submission history
From: Daniel Nevo [view email][v1] Thu, 2 Apr 2015 09:44:17 GMT (64kb,D)
[v2] Sun, 26 Jul 2015 17:09:13 GMT (65kb,D)
[v3] Wed, 25 Nov 2015 11:28:38 GMT (66kb,D)
Link back to: arXiv, form interface, contact.