Transfer Learning Technique: An Approach For Network Intrusion Detection

Abstract

Transfer learning (TL) is a machine learning technique that applies knowledge gained from solving one problem to solve a different but related problem. TL reduces the effort and time required to solve a problem by using a pre-trained model on a similar problem. Recently, network intrusion detection systems have incorporated both traditional and deep learning techniques. The goal of these systems is to detect network attacks and violations of rules. Although machine learning introduces novel attack surfaces that fascinate researchers, it also challenges the development of novel models, input definitions, and ML-IDS training. This research investigates the application of transfer learning as a technique for network intrusion detection. The proposed IDS system is capable of detecting multiple threat levels in the networking field. The system uses two pre-trained convolutional neural networks (CNNs) from the perspective of knowledge transfer, combining their predictions into a single model. The pre-trained CNNs on a related problem are adapted for a new but similar problem. The models were trained and tested using transfer learning to detect both common and uncommon attack patterns. The NSL-KDD dataset from Kaggle was employed to test the network intrusion detection system. The experimental results demonstrate that our TL technique has a prediction accuracy of 96.52%, which is a remarkable level of efficiency as expected. We therefore recommend the application of transfer learning, which has greatly contributed to the development of deep learning and is essential in developing efficient network intrusion detection systems