Deep Reinforcement Learning for Adaptive Cyber Defense and Attacker’s Pattern Identification

Abstract

The cyberattack has been increased rapidly since most of the devices are connected to networks due to Internet of Things (IoT) era. In addition, artificial intelligence (AI) is booming due to its capability to adapt to most of the science fields. AI has the capability to learn, identify, and recognize certain pattern according to their training approaches. This paper aims to propose reinforcement learning for adaptive cyber defense that is capable to avoid a certain pattern of attack and identify the pattern of attack from the cyber outlaw. The algorithm will learn and identify the behavior of the attack and attackers through the training dataset and then provide a counterattack to avoid unnecessary loss. The experimental result has shown the fitness of neural network algorithm and the proposed reinforcement learning framework with a 95% confidence rate. The correct prediction from the confusion matrix has also shown high value with 96%. The future works will be focused on the real-world data testing and hard-coded reinforcement algorithm to observe adaptability of the proposed framework.