Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 357 - Contemporary Multivariate Methods
Type: Contributed
Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #313090
Title: Model Selection Criteria for Biological Networks by Using Loop-Based Multivariate Regression Adaptive Splines Model
Author(s): Gul Bulbul* and Vilda Bahar Purutçuo?lu
Companies: Bowling Green State Univ and Middle East Technical University
Keywords: High dimensional data; StARS; GCV; biological networks; high dimensional model selection; LMARS

The static biological networks are known for their sparsity and complexity. In the statistical literature, the undirected graphical models provide a useful way to describe their nature in higher dimensions. Loop-based Multivariate Adaptive Regression Splines (LMARS), which is one of the strong nonparametric approaches for modeling these complex networks as the alternative of these approaches such as Gaussian graphical model, copula GGM and clustering-based methods. Basically, LMARS uses generalized cross-validation as its standard model selection technique like Akaike and Bayesian information criteria in its calculation while detecting the optimal description of the network. However, GCV cannot work properly for higher dimensional biological networks. Thus, we suggest applying Stability Approach to regularization Selection which is a stability-based method for finding the best model by using a random subsampling procedure, as a regularization parameter selection technique within LMARS in place of GCV. Thereby, in this study, we compare the performances of StARS with GCV in LMARS under different dimensional and topological structures while using both real and simulated datasets.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program