DDoS Attack Detection using Supervised Machine Learning Techniques

Computer Science Engineering

There has been a massive growth in cyberspace in the last few decades which has demanded the implementation of efficient networks for communication. There are various types of cyber-attacks, and they grow in congruence with every new technology where each has a varying level of threat impact. In a simple network sniffing attack, the target might not experience any consequences while the impact of a DDoS attack is on the contrary. DDoS attacks are simple to perform but the effects of it result in production downtime, financial losses, customer dissatisfaction and loss of reputation of the targeted businesses therefore becoming an important security concern. This study proposes an effective solution for the detection of DDoS attacks using four different Supervised Machine Learning techniques including Random Forest, K-Nearest Neighbor, AdaBoost and Logistic Regression. These algorithms are trained and tested with a subset of the CICDoS 2019, 2018 and 2017 datasets; and the classification accuracy scores are 99.99%, 99.92%, 99.97% and 98.06% respectively. The dataset cluster consists of about 5,00,000 records and 8 features were selected as necessary from a total of 80 features. The training time and data class classification metrics are considered for the determination of the most appropriate Supervised ML technique for the base model and Random Forest is selected as the base for validation since it performs better when compared to other techniques. Wireshark is used to gather real time network traffic and the captured packets are then given as input for the trained Random Forest model for validation purpose and it is found that the prediction values are accurate. The prediction '0' suggests that the nature of the traffic is Benign and the predicted '1' suggests that the nature of the traffic is DDoS.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.