DDoS detection using Machine Learning

This project intends to detect DDoS attacks with machine learning techniques. Currently, Stream Random Patches with Hoeffding Trees are being explored.

Dependencies are:

river==0.15.0

numpy==1.24.2

pandas==1.5.3

Next steps

• Change plotting: calculate statistics in all data used in training so far
• Change shuffling: it's interesting to keep data in its real order. It may worsen the results, but it's closer to the real word situation
• Try AdaCost: to be able to assign misclassification costs may improve results.
• Try no ensemble: are the ensembles doing anything?
• Run new models on older datasets. It's importante to have data for comparison.
• Try new base estimators: other trees, SVM's (?), NN's (?), etc.
• Try to consume dataset by windows: it may be useful to extract features of windows (by time or number of packets). It's faster and may be more accurate.

Next meeting at: 12/04/2023

Next presentation at: 26/04/2023

Interesting article: https://ieeexplore.ieee.org/document/6488798 Interesting: