Algorithm 1 Greedy Algorithm for Feature Selection

Input: Features and labels for dysarthric and healthy speakers: XD, yD, XC, yC

Output: Top k features that optimize criteria in (7): Ω

Define: $f(\mathrm{\Omega })=\frac{{max}_{\theta }{l}_D(\theta |{\mathbf{X}}_D(\mathrm{\Omega }),{\mathbf{y}}_D)}{{max}_{\zeta }{l}_C(\zeta |{\mathbf{X}}_C(\mathrm{\Omega }),{\mathbf{y}}_C)}$

Ω = ∅

F = 1 … M

for j ∈ 1 … k do

J = ∅

for Fi ∈ F \ Ωj do

J(Fi) = f(Ω ∪ Fi)

end for

$$\mathrm{\Omega }=\mathrm{\Omega }\cup \{\underset{F_i}{\mathbf{argmax}}J(F_i)\}$$

end for