An Advanced DDOS Attack Detection Model with an Ensembled SVM and Baruta Selection Technique

Abstract

The paper proposed the use of an ensembled SVM model with the Boruta selection techniqueto improve cloud DDoS attack detection. DDoS attacks are the most common cloudsecurityattacks, with a 16% level of use. They can render the entire system useless, with resourcesoffline for 24 hours, multiple days, or even a week depending on the severity of the attack. Inthe event of successful attacks, about $ 20,000 can be lost by a company. DDoS attacks canalso make the cloud environment vulnerable to hacking, due to bad hosting or sharedhosting, failure to prepare against the attack, outdated codes, and other issues. This studyaimstoimprove the performance of Support Vector Machine (SVM) to better detect CloudDDoSattacks by eliminating key problems and improving memory efficiency, effectiveness, andhigh dimensional space. Several Machine learning techniques like Decision Tree, RandomForest, KNN, and SVM were used to detect DDoS attacks in a cloud environment. In terms of detection accuracy SVM is the best among the used techniques with 84.94%. Aproposed ensembled SVM with the Boruta selection technique was modeled to improve the performanceof DDoS attack detection techniques in the cloud. Five different models were designedusingdistinct machine-learning techniques and compared to the proposed model for better performance. Logistic regression, Random Forest Classification, Support Vector Machine, K-Nearest Neighbor, and Linear Discriminant Analysis. All five Classifiers wereusedindependently and with the Bagging technique, giving different results in all aspects. Fromtheir performance found that after the boruta selection extract 51 features out of the 79 original features of the and the  data that was summed up to 1048575 was reduced to 1025 for optimal performance, Random Forest Classifier and K-Nearest Neighbor  was said to performbetterthan the proposed SVM classifier in both Individual modeling and with Bagging Ensembled learning. A great  improvement was achieved by the model performance with a detection accuracy of 95.7%, 10.8% more than the traditional SVM, an  improvement the accuracy. Theimplementation of KNN, Random Forest, and Linear Discriminant analysis in ensembledlearning shows  that their performance is better than the proposed system.