Figure 7
Boxplots of QDA performance. Scenarios 3 and 4. The figure displays boxplots of the QDA 10-fold cross validation error rate for 100 simulations of the following experiment: 80 observations, with class sizes n0 = n1 = 40, were generated for different dimensions p = 2, 5, 10, 15,20, 30 of the feature space for chunks of noisy variables and once again for chunks containing the weak marginal/strong bivariate signal. The QDA error rate is computed in both cases. Boxplots in blue correspond to the cv QDA error rate for chunks with the weak marginal/strong bivariate signal; on the other hand, boxplots in pink correspond to the cv QDA error rate for blocks containing only noisy features. The amount of overlap between both populations of boxplots is shown in parenthesis. The simulation was carried out for the weak marginal/strong bivariate signals of scenarios 3 and 4.