Dynamic Packet Filtering Using Machine Learning

With the advent of the Internet, cyber-attacks and threats have become a major concern. Traditional methods of manual network monitoring and rule-based packet filtering are tedious and have become less effective against attacks. Filtering packets purely based on payload and pattern matching are also inefficient. There is need for a dynamic model which can learn the rules to filter packets. This paper proposes a machine learning-based packet filtering model using Neural Networks. After developing a classified model with training and validation data, the model can be utilized to support dynamic packet filtering. The proposed model provides the capability to filter the packets not purely based on static rule-based filtering, but on attributes in IP packets and previously learned rules from the model. The proposed model considers both payload and other attributes in the IP packet for filtering. The model can automatically update the firewall rules to enhance security.