Fig 8. Average AUC scores over the four classifiers using ranked events.

The events were ranked using t-test statistic (t-test), mutual information (MI) and chi-squared testing (chi²). The horizontal line shows the performance when including all features (i.e. without ranking). 100 means including all informative events, i.e. events with a rank above zero, 50 means including 50% of all features with a rank above zero. Including the top 10% of all events was most effective for all classifiers except SVM for which using the top 1% was the most effective choice.