Network Intrusion Detection using Deep Reinforcement Learning

The number of internet-connected systems has increased by a huge amount in recent years. These systems are highly vulnerable and increasingly at risk from cyber- attacks. These cyber-attacks need more advanced and intelligent cyber defense systems that include agents that can learn to make judgements on their own without consulting human specialists. Deep learning-based reinforcement learning (DRL) is very good at handling complicated, dynamic, and especially high-dimensional cyber protection problems. This article provides a modern network intrusion detection technology, which employs a deep feed-forward neural network method and reinforcement learning, which is based on Q- learning. In order to detect various sorts of intrusions in the network using an automated trial- and-error method and continually improve its detection skills, the Deep Q-Learning (DQL) model is proposed. The accuracy of the model proposed is 91.4%, while the accuracy of other self-taught learning models is 88.4% and it is a similar case for recall rate and precision as well which are 90.2% and 92.8%. Our experimental findings further demonstrate that our suggested DQL beats other comparable machine learning methods and is very good at identifying various intrusion classifications.