An Optimal Reinforced Deep Belief Network for Detection of Malicious Network Traffic

Abstract

Network traffic analysis is referred to as the method to monitor the availability of the network and identify the anomalous activities like operational issues and security. Network traffic mainly occurs due to the transmission of a large amount of data over the computer network at the same time. The grouping and classification process can evaluate the network traffic and the endless determination of network traffic patterns is the most important challenge during network traffic classification. The existing approaches have some limitations because the time consumption was not able to reduce and it has low classification accuracy in the network traffic analysis. To overcome these problems, the Dove Swarm Optimized Reinforced Deep Belief Network (DSO-RDBN) is proposed which is utilized for optimizing the hyperparameters. There are no effectual training algorithms that can optimize the different hyperparameters associated with the RDBN architecture such as momentum coefficients, learning rate, number of hidden nodes, and number of hidden layers without manual tuning. The manual selection of hyperparameters leads to some delay in the network traffic classification process so the DSO algorithm is implemented for addressing the hyperparameter tuning problem. For network traffic classification analysis, the USTC-TFC2016 test dataset is used and this dataset has two categories Benign and Malicious. The performance metrics such as accuracy, precision, sensitivity, F-measure, and false alarm range (FAR) are applied for improving the performance compared to state-of-art methods. The performance rate of accuracy, precision, sensitivity, F-measure and FAR is 97.2%, 97.6%, 97.5%, 97.6% and 68% respectively.