DDoS attacks are type of attack that the most common and very massive lately. To overcome this attack, we need an appropriate mechanism that can handle it. One way is to resolve that problems are use ML assistance. This paper will be describing the implementation of DDoS classification using the ML Semi-Supervised Learning Pseudo Labeling algorithm, which that algorithm will be running on a computer device that pre-installed MHN-sensor before. The MHN-sensor will function as a fake server that filter out DDoS data, and then will be processed by our program. The program we have created will running the Semi-Supervised Learning Pseudo Labeling algorithm which aims to classify Normal and DDoS traffic data. This program runs by combining Supervised and Semi-Supervised algorithms, which are algorithms that work with a data clustering mechanism that works twice, so the ML processing will become be better. Other than that, we also carry out CPU-usage testing which aims to determine the effectiveness of our program performance. Algorithm testing that has been used in this research program shows excellent results. For testing the program shows an average result of 98.8%, meanwhile when the program is tested in real time on the SDN network it shows an average result of 60%. And for CPU usage it is relatively small, which is only 14% around. As for the further development of this test, namely adding a mitigation feature by utilizing the SDN network. In the future, when there is an incoming attack, the program will give commands to the controller to mitigate and temporarily stop incoming traffic.
Keyword: Machine Learning, DDos, SDN, MHN, Semi-Supervised Learning.
text is fully visible on http://eprints.umm.ac.id/70054/
full view poster image at https://iili.io/KeT3zB.jpg
