A Comparison of Penalized Regressions for Estimating Directed Acyclic Networks

Abstract

Network models can be classified into two large groups: undirected and directed. Directed network graphs that can represent causal relationships are likely more appropriate in bio-medical data. There have been many studies to estimate DAGs(Directed Acyclic Graphs), of which the two-stage approach using lasso effectively. Find the edges between the nodes in the first step and find the direction in the second step. In this paper, we try to compare which penalized regression is better to find neighborhoods through simulations. We present the result of the simulations that shows which penalized regression is the best.