Özalp, A.N.Albayrak, Z.2024-09-292024-09-2920221785-8860https://doi.org/10.12700/APH.19.7.2022.7.12https://hdl.handle.net/20.500.14619/9201In computer networks, intrusion detection systems are used to detect cyberattacks and anomalies. Feature selection is important for intrusion detection systems to scan the network quickly and accurately. On the other hand, analyzes performed using data with many attributes cause significant resource and time loss. In this study, unlike the literature studies, the frequency effects of the features in the data set are analyzed in detecting cyber-attacks on computer networks. Firstly, the frequencies of the features in the NSL-KDD data set were determined. Then, the effect of high-frequency features in detecting cyber-attacks has been examined with the widely used machine learning algorithms of Random Forest, J48, Naive Bayes, and Multi-Layer Perceptron. The performance of each algorithm is evaluated by considering Precision, False Positive Rate, Accuracy, and True Positive Rate statistics. Detection performances of different types of cyberattacks in the NSL-KDD dataset were analyzed with machine learning algorithms. Precision, Receiver Operator Characteristic, F1 score, recall, and accuracy statistics were chosen as success criteria of machine learning algorithms in attack detection. The results showed that features with high frequency are effective in detecting attacks. © 2022, Budapest Tech Polytechnical Institution. All rights reserved.eninfo:eu-repo/semantics/openAccessAnomaly detectionAttribute selectionCyberattacksIDSMachine LearningNSL-KDDArticle10.12700/APH.19.7.2022.7.122-s2.0-851387485822337Q221319