Effectively predicting cyber‐attacks through isolation forest learning‐based outlier detection

Abstract

AbstractDue to the popularity of Internet of Things devices, the exponential progress of computer networks, and a plethora of associated applications, cybersecurity has recently attracted much attention in light of today's security problems. As a result, detecting various cyber‐attacks within a network and developing an effective cyber‐attacks prediction model that plays a crucial part in today's defense has become increasingly critical. Modeling cyber‐attacks effectively, on the other hand, is challenging because modern security datasets hold a large number of dimensions of security features and may contain outliers. To accomplish this, we provide an approach for categorizing cyber‐attacks effectively through isolation forest learning‐based outlier detection. Additionally, we apply a variety of popular machine learning approaches to assess the performance of cyber‐attacks prediction models, including logistic regression, support vector machine, AdaBoost classifier, naive Bayes, and K‐nearest neighbor. We evaluated the efficacy of our approach by running tests on three network intrusion datasets (KDD Cup 99, CIC‐IDS2017, and UNSW‐NB15) and computing the precision, recall, and accuracy. Experiments demonstrate that eliminating outliers improves the prediction accuracy of cyber‐attacks for different classifiers. Additionally, we compare the isolation forest learning‐based outlier detection model to other well‐known outlier detection techniques, DBSCAN and k‐means, and measure the effectiveness of our model.