Variable Selection in Multivariate Multiple Regression

Recently, we have tackled the problem of unsupervised binary feature selection by proposing a sta- tistical framework based on finite multivariate Bernoulli mixture models which

Models in competition are composed of the relevant clustering variables S, the subset R of S required to explain the irrelevant variables according to a linear regression, in

Literature on this topic started with stepwise regression (Breaux 1967) and autometrics (Hendry and Richard 1987), moving to more advanced procedures from which the most famous are

Introduced by Friedman (1991) the Multivariate Adaptive Regression Splines is a method for building non-parametric fully non-linear ANOVA sparse models (39). The β are parameters

By taking account of model uncertainty in quantile regression, there is the possibility of choosing a single model as the “best” model, for example by taking the model with the

Table 3 provides estimates of the model parameters for the median case (τ = 0.5) as well as a comparison with the results obtained by three different estimation approaches, namely

We found that the two variable selection methods had comparable classification accuracy, but that the model selection approach had substantially better accuracy in selecting

Keywords and phrases: Categorical multivariate data, clustering, mix- ture models, model selection, penalized likelihood, population genetics, slope heuristics,