Algorithms for Sparse Support Vector Machines

Many problems in classification involve huge numbers of irrelevant features. Model selection reveals the crucial features, reduces the dimensionality of feature space, and improves model interpretation. In the support vector machine literature, model selection is achieved by ℓ1 penalties. These convex relaxations seriously bias parameter estimates toward 0 and tend to admit too many irrelevant features. The current paper presents an alternative that replaces penalties by sparse-set constraints. Penalties still appear, but serve a different purpose. The proximal distance principle takes a loss function L(boldsymbolβ) and adds the penalty fracρ2dist(boldsymbolβ, Sk)2 capturing the squared Euclidean distance of the parameter vector boldsymbolβ to the sparsity set Sk where at most k components of boldsymbolβ are nonzero. If boldsymbolβρ represents the minimum of the objective fρ(boldsymbolβ)=L(boldsymbolβ)+fracρ2dist(boldsymbolβ, Sk)2, then boldsymbolβρ tends to the constrained minimum of L(boldsymbolβ) over Sk as ρ tends to ∞. We derive two closely related algorithms to carry out this strategy. Our simulated and real examples vividly demonstrate how the algorithms achieve much better sparsity without loss of classification power.