Fig. 18
From: Hidost: a static machine-learning-based detector of malicious files
Feature set drift in the PDF dataset. Illustrates change in the feature set between consecutive feature selection events in our experiment, performed before every training, i.e., once per week. For weeks 2 to 10, the percentage of features that have been added to (new), removed from (obsolete), or remained (unchanged) in the feature set is plotted, relative to the previous week