Open AccessTCP/UDP-Based Exploitation DDoS Attacks Detection Using AI Classification Algorithms with Common Uncorrelated Feature Subset Selected by Pearson, Spearman and Kendall Correlation Methods
Author(s) -
Kishore Babu Dasari,
Nagaraju Devarakonda
Publication year - 2022
Publication title -
revue d'intelligence artificielle
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.146
H-Index - 14
eISSN - 1958-5748
pISSN - 0992-499X
DOI - 10.18280/ria.360107
Subject(s) - denial of service attack , random forest , computer science , artificial intelligence , naive bayes classifier , pearson product moment correlation coefficient , machine learning , correlation , data mining , multilayer perceptron , feature selection , classifier (uml) , intrusion detection system , pattern recognition (psychology) , artificial neural network , support vector machine , mathematics , statistics , the internet , geometry , world wide web
The Distributed Denial of Service (DDoS) attack is a serious cyber security attack that attempts to disrupt the availability security principle of computer networks and information systems. It's critical to detect DDoS attacks quickly and accurately while using as less computing power as possible in order to minimize damage and cost efficient. This research proposes a fast and high-accuracy detection approach by using features selected by proposed method for Exploitation-based DDoS attacks. Experiments are carried out on the CICDDoS2019 datasets Syn flood, UDP flood, and UDP-Lag, as well as customized dataset. In addition, experiments were also conducted on a customized dataset that was constructed by combining three CICDDoS2019 datasets. Pearson, Spearman, and Kendall correlation techniques have been used for datasets to find un-correlated feature subsets. Then, among three un-correlated feature subsets, choose the common un-correlated features. On the datasets, classification techniques are applied to these common un-correlated features. This research used conventional classifiers Logistic regression, Decision tree, KNN, Naive Bayes, bagging classifier Random forest, boosting classifiers Ada boost, Gradient boost, and neural network-based classifier Multilayer perceptron. The performance of these classification algorithms was also evaluated in terms of accuracy, precision, recall, F1-score, specificity, log loss, execution time, and K-fold cross-validation. Finally, classification techniques were tested on a customized dataset with common features that were common in all of the dataset’s common un-correlated feature sets.