Bi-directional Recurrent Neural network for Intrusion Detection System (IDS) in the internet of things (IoT)

With IoT technology bringing a large number of day-to-day objects into the digital fold to make them smarter. It is also evident that the IoT technology is going to transform into a multi-trillion-dollar industry in the near future. However, the reality is that IoT bandwagon rushing full steam ahead is prone to countless cyber-attacks in the extremely hostile environment like the internet. Nowadays, standard Personal Computer (PC) security solutions won't solve the challenge of privacy and data security transmitted over the internet. In this research, we have applied a Bidirectional Recurrent Neural Network (BRNN) to build a security solution with high durability for IoT network security. Deep learning and Machine learning have shown remarkable result in dealing with multimodal and voluminous heterogenous data in regards to intrusion detection especially with the architecture of Recurrent Neural Network(RNN). Feature selection mechanisms were also implemented to help identify and remove non-essential variables from data that does not affect the accuracy of the prediction model. In this case a Random Forest (RF) algorithm was implemented over Principal Component Analysis (PCA) because of flexibility, and easy in using machine learning algorithm that allows production without hyper-parameter tuning, building of multiple decision trees and merging them together to get a more accurate and stable prediction. In this study a novel algorithm (BRNN) out-performed both RNN(Recurrent Neural Network) and GRNN(Gated Recurrent Neural Network) because it considers both information from the past and the future with back and forward hidden neurons.