Anomaly based network intrusion detection using machine learning

The online dependency of the world has been sharply increasing for the last two decades, and this year, due to the pandemic, all sorts of life literally moved to online. Parallel to these, the number of attacks and criminal activities on the internet is steadily rising. Several security systems are deployed to protect network infrastructure against criminals. Network Intrusion Detection System (NIDS) is one of the most popular security systems used. It can be a signature-based or anomaly-based system. Although signature-based NIDS are efficient in preventing known attacks, they are futile against zero-day attacks. In detecting new attacks, the anomaly-based method is proved to be efficient. In this study, machine learning techniques are used for network anomaly detection. To this effect, a bunch mark dataset, CSE-CIC-IDS2018 is used. It is pre-processed and important features selected with Random Forest Regressor algorithm. The cleaned dataset with the selected features is fed to eight different machine learning algorithms. After reducing the number of features from the original 80 features to 17 features, the machine learning algorithms achieved the following success rates: Naïve Bayes 36%, QDA 50%, Random Forest 94%, ID3 94%, AdaBoost 94%, MLP 77%, and K Nearest Neighbours 95% and Gradient Boosting 95%.