Fig. 4

Average and worst case performance of feature selection algorithms. Average and worst case performance are obtained using 27 combinations of the synthetic microarray model parameters k, ρ₀ and ρ₁ with fixed mean type, where performance is defined to be the average number of markers identified as good plus the average number of non-markers identified as bad over 500 iterations: a average performance for redundant means, b worst case performance for redundant means, c average performance for synergetic means, d worst case performance for synergetic means, e average performance for marginal means, and f worst case performance for marginal means