Federated Intrusion Detection System using Liquid Neural Network

Abstract

In the field of cybersecurity, Intrusion Detection Systems (IDS) play a crucial role in preventing unauthorized access, use, or modification of computer systems and networks. IDS can be classified into two types based on their detection method: signature-based and anomaly-based detection. The former detects known signatures of attacks in the incoming traffic, while the latter identifies deviations from normal traffic by learning its behavior. Deep Learning (DL) techniques such as Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs) have shown promise in improving the accuracy of IDS. In this research, we assess the performance of various DL models on two widely used datasets, NSL KDD and CSE-CIC-IDS2018. Also, it includes implementation of Federated learning algorithm for Intrusion detection systems with 2 clients and one server. Our findings indicate that LNN can predict data more accurately, while LSTM and GRU can better train on the data. Furthermore, LNNs results in 98.63% accuracy, 98.60% Precision, and 97.88% recall. In conclusion, LNNs work at par with LSTMs and GRU with the fact that LNNs are fairly new and further tunings can help in getting better results. For Federated algorithm, LNN was used as base model which brings up to 98.5% of accuracy.