Iterative Feature Selection-Based DDoS attack Prevention Approach in Cloud

Abstract

Distributed Denial of Service (DDOS) attacks aim to exploit the capacity and performance of a network's infrastructure, making the cloud environment one of the biggest targets for attackers. Many efforts are being made in the field of technology to prevent them from disrupting the services provided. Machine Learning techniques are a means to protect against DDOS attacks. Data preprocessing, feature selection, and classifiers are the main components of any prevention framework. The focus of this study is to find and enhance the feature selection approach for increasing the accuracy of the classifiers in detecting DDOS attacks from regular traffic. We used four different techniques, including Pearson Correlation Coefficient (PCC), Random Forest Feature Importance (RFFI), Mutual information (MI), and Chi-squared(X2) measure which we tested on different classifiers. The first selection approach was based on the feature’s independency level then the second iteration was based on the feature’s importance. We also examined the claim of dropping attacks from the dataset for better accuracy. The best performing set of features was from using PCC and RFFI together for feature selection with average accuracy and precision of 99.27% and 97.60%, which is higher than the use of PCC for both measures by almost 2%. The accuracy is also higher by nearly 12% from the same approach dropping 50% of the attacks.