Improvement Of An Intrusion Detection System Based On Deep Belief Networks Models: A Review

Abstract

Technology is rapidly evolving in a world powered by social networks, online transactions, cloud computing, and automated processes.  However, as technology develops, equal  cybercrime.  Cyber attacks are increasing rapidly, making cybersecurity a challenge in the digital era.  Intrusion detection systems (IDS) are an advancement that enhances  network security and protects an organization's data.  IDS helps  network administrators  detect  malicious activity within the network and alerts administrators to protect data  by taking  appropriate measures against these attacks.  Deep Belief Networks (DBN) are generative graphics models formed by stacking multiple Restricted Boltzmann Machines (RBMs).  High-dimensional representations can be identified and learned.Improving and evaluating Deep Belief Networks (DBN) for detecting cyber-attacks in a network of connected devices using the CICIDS2017 dataset. Several class balancing techniques were aplied and evaluated. The recomendation to improve IDS based on DBN is collect more data, increase the number of layers, tune the hyperparameters, regularize the network, and use more efficient training algorithms.