A Search for Computationally Efficient Supervised Learning Algorithms of Anomalous Traffic

Abstract

Recently, due to the growing use of the Internet of Things (IoT) and mobile networks, Internet traffic has been rapidly growing. Information security is a serious problem due to a variety of intrusion incidents of the Internet and various types of network attacks. Since most commercial products of network-based intrusion detection systems which are currently used utilise expert-based misuse detection techniques and statistic-based anomalous behaviour detection techniques, these techniques are still too limited to completely detect various types of network attacks. Using the KDD Cup 1999 data set, well-known supervised learning algorithms of many machine learning algorithms to automatically generate knowledge under our proposed and implemented detect system are applied for normal network packets and various anomalous network packets in this paper. Based on such learned knowledge, experiments to determine whether it is normal or abnormal are examined for various packets, and accuracy and processing speed for five selected supervised learning algorithms are compared and analysed. As a result of analysing the accuracy and processing speed of five well - known supervised learning algorithms, SVMWithSGD and Logistic Regression algorithms have been determined to show the most accurate results. With regard to processing speed, Random Forest and Decision Tree algorithms are the fastest algorithms.