Exploring Deep Neural Network Capability for Intrusion Detection Using Different Mobile Phones Platforms

Abstract

Network intrusion activities started since the network typologies become available for public, the target devices were computers, servers, switches, routers and so on, but with the fever of smartphones spreading among the public, these smartphones have become a target for hacker attacks. Network Intrusion Detection System (IDS) is a device, software or both used by network administrator to monitor the network security and recognize any malicious activity or organization's network policy violation. Deep neural network is the famous technology in deep learning where it is an artificial neural network with multiple hidden layers, and it simulates the human brain and it's the central nervous system in thinking and detecting pattern. Mobile phones are widely used today, with different platforms which known now as smartphones, these smartphones belong to many manufacturing and require operating system. This research explores deep neural networks capability for intrusion detection using different mobile phones platforms, the research experiments five deep neural networks approaches, Fully Connected Deep Neural Networks (FCDNN), Convolutional Neural Networks (CNN), Convolutional Neural Networks followed by Fully Connected Neural Networks (CNN&FCDNN), Convolutional Neural Networks followed by Fully Connected Neural Networks and then followed by Convolution Neural Network (CNN&FCDNN&CNN), and finally Fully Connected Deep Neural Networks followed by Convolution Neural Networks (FCDNN&CNN).Two data sets have been used for testing the approaches, Android data set and NSL-KDD data set. The proposed approaches have been compared with each other's and also compared with traditional Machine Learning (ML) approaches. The results show that CNN&FCDNN&CNN is the best deep learning approach where it achieved accuracy of 0.863 and 0.997 for the two data sets Android and NSL-KDD respectively. The accuracy results also show that the best approach is the random forest where it achieved 0.88 and 0.998 for Android and NSL-KDD data sets respectively. Deep neural networks show that they are good machine learning candidates for problems similar to mobile phones intrusion detection systems.