Abstract:
With the increasing occurrence and complexity of cyber-threats, there is a need today
to make use of highly performant and robust detection tools for advanced
cyber-attacks such as Distributed Denial of Service (DDoS) attacks in the rapidly
changing cyber-space-centric environment. In this paper, we propose a new ensemble
approach that combines different machine learning models to improve the detection
and classification of different DDoS attacks. We experimentally evaluate the proposed
ensemble models for DDoS detection on a large and diverse but also imbalanced
dataset of DDoS attack instances drawn from the CIC-DDoS2019. In this work, we
analyze the influence of different ensemble voting classifiers to see the effects on the
performance of the final models. Results show that 99.81% accurate detection class
and 93.92% accurate classification class can be determined by our ensemble approach
for identifying and classifying DDoS attacks which reflect the effectiveness and
robustness of our method. We used evaluation metrics like validation accuracy and
confusion matrices for the results revealed that our model is very interesting to detect
cybersecurity problems such as DDoS. We are also writing a web app to plug in these
performant detection, and recognition algorithms. By utilizing this platform, cyber
security professionals around the world can have monitoring of their system logs and
even be empowered to take necessary preemptive steps, thus creating an environment
of heightened awareness and readiness for overall cyber security. In this paper, we are
not only able to achieve a high detection and classification rate in DDoS attack
detection and recognition by the proposed ensemble-based model, but also construct
the premise of network security in academic research and network application
especially when it comes to log file analysis.
