A modified score function estimator for multinomial logistic regression in small samples
Computational Statistics & Data Analysis
Fast and Robust Smallest Enclosing Balls
ESA '99 Proceedings of the 7th Annual European Symposium on Algorithms
Statistical and Inductive Inference by Minimum Message Length (Information Science and Statistics)
Statistical and Inductive Inference by Minimum Message Length (Information Science and Statistics)
Invariance priors for Bayesian feed-forward neural networks
Neural Networks
| Hi-index | 0.00 |
Parameters in logistic regression models are commonly estimated by the method of maximum likelihood, while the model structure is selected with stepwise regression and a model selection criterion, such as AIC or BIC. There are two important disadvantages of this approach: (1) maximum likelihood estimates are biased and infinite when the data is linearly separable, and (2) the AIC and BIC model selection criteria are asymptotic in nature and tend to perform well only when the sample size is moderate to large. This paper introduces a novel criterion, based on the Minimum Message Length (MML) principle, for parameter estimation and model selection of logistic regression models. The new criterion is shown to outperform maximum likelihood in terms of parameter estimation, and outperform both AIC and BIC in terms of model selection using both real and artificial data.