Figure 4

A 10-fold cross validation of the original data. When analyzing non-simulated data, LDA and QDA show a weaker performance than in the simulated data sets. PDA also suffers a bit, but still achieves a very high accuracy. The random partitioning of the data into 10 folds was performed 50 times, and the resulting standard deviations over 50 repetitions are shown in the error bars.