Scale-Invariant and consistent Bayesian information criterion for order selection in linear regression models
Abstract
This paper presents a new form of the Bayesian information criterion (BIC) named as BIC-robust for order selection in linear regression models. Original BIC is not consistent for fixed sample size and high signal-to-noise-ratio (SNR) cases. Also, the high-SNR forms of the BIC suffer from data scaling problem. BIC-robust eliminates this scaling problem, hence it is scale-invariant. It is also a consistent estimator of the true model order for large sample size and high-SNR scenarios.
- Publication:
-
Signal Processing
- Pub Date:
- July 2022
- DOI:
- 10.1016/j.sigpro.2022.108499
- Bibcode:
- 2022SigPr.19608499G
- Keywords:
-
- BIC;
- Consistency;
- Linear regression;
- Model order selection;
- Scale-invariant