A Comparative Analysis of Machine and Deep Learning Classifiers for Intrusion Detection

Presently available security tools are eluded by polymorphic malware, zero-day vulnerabilities, and unauthorized attempts. These open up loophole points in the systems for intruders to access the private network as legitimate users. An adversary can use similar commands, scripts, tools, and credentials as a system administrator to make it impossible to find the difference between a legitimate user and the intruder. A good classifier aims to classify such events automatically based on learning strategies. This study aims to analyze the appropriate classifier for the use of the Industrial Internet of Thing (IIoT). Some of the well-known classifiers from the domain of machine and deep learning techniques are experimented. The results are compared to evaluate the model applicable to industrial devices and provide appropriate security. This study utilizes two popular datasets: the UNSWNB-15 and KDD+ datasets to experiment with the models. Finally, a comparative analysis is evaluated based on accuracy, false rate, and time is taken for detection. Deep learning models claim better efficiency and thus, are suitable for IIoTs significantly.