Efficiency of Different Machine Learning Algorithms on the Multivariate Classification of IoT Botnet Attacks

The Internet of Things, with its enormous growth in the recent decades, has not just brought convenience to the different aspects of our lives. It has also increased the risks of various forms of cybercriminal attacks, ranging from personal information theft to the disruption of the entire network of a service provider. As the demands of such devices increase rapidly on a global scale, it has become increasingly difficult for different corporations to focus on security efficiently. As such, the demand for methodologies that can aptly respond to prevent intrusion within a network has soared disturbingly. Various utilization of anomaly traffic detection techniques has been conducted in the past, all with the similar aim to prevent disruption in networks. This research aims to find an efficient classifier that detects anomaly traffic from N_BaIoT dataset with the highest overall precision and recall by experimenting with four machine learning techniques. Four binary classifiers: Decision Trees, Extra Trees Classifiers, Random Forests, and Support Vector Machines are tested and validated to produce the result. The outcome demonstrates that all the classifiers perform exceptionally well when used to train and test the anomaly within a single device. Moreover, Random Forests classifier outperforms all others when training is done on a particular device to test the anomaly on completely unrelated devices.