Tight conditions for consistency of variable selection in the context of high dimensionality

Laëtitia Comminges (1, 2) , Arnak S. Dalalyan (1, 2)

1 LIGM - Laboratoire d'Informatique Gaspard-Monge
2 imagine [Marne-la-Vallée]

We address the issue of variable selection in the regression model with very high ambient dimension, \textit{i.e.}, when the number of variables is very large. The main focus is on the situation where the number of relevant variables, called intrinsic dimension and denoted by $d^*$, is much smaller than the ambient dimension $d$. Without assuming any parametric form of the underlying regression function, we get tight conditions making it possible to consistently estimate the set of relevant variables. These conditions relate the intrinsic dimension to the ambient dimension and to the sample size. The procedures that are provably consistent under these tight conditions are simple: they are based on comparing the empirical Fourier coefficients with an appropriately chosen threshold value. The asymptotic analysis reveals the presence of two quite different re\-gimes. The first regime is when $d^*$ is fixed. In this case the situation in nonparametric regression is the same as in linear regression, \textit{i.e.}, consistent variable selection is possible if and only if $\log d$ is small compared to the sample size $n$. The picture is completely different in the second regime, $d^*\to\infty$ as $n\to\infty$, where we prove that consistent variable selection in nonparametric set-up is possible only if $d^*+\log\log d$ is small compared to $\log n$.

Mots clés: variable selection; nonparametric regression; sparsity pattern

Domaines: Statistiques [math.ST]; Théorie [stat.TH]; Machine Learning [stat.ML]

Laëtitia Comminges, Arnak S. Dalalyan. Tight conditions for consistency of variable selection in the context of high dimensionality. 2011. ⟨hal-00602211v1⟩