Fares Meghdouri, Félix Iglesias Vázquez, Tanja Zseby
Fares Meghdouri, Maximilian Bachl, Tanja Zseby
Maximilian Bachl, Fares Meghdouri, Joachim Fabini, Tanja Zseby
Fares Meghdouri, Tanja Zseby, Félix Iglesias Vázquez
I analyze the effect of the observation time on network attack detection performance since attacks are of different duration and a fixed length time window seems to be a bad idea.
Similarly to Random Forest, I propose an algorithm that creates a decision tree per class and builds the final classifier by performing majority guessing.
I propose a new algorithm for reverse engineering feature importance of trained models.
The algorithm serves to improve accuracy and generate adversarial samples for defense purposes.