Deep Transfer Learning for IoT Intrusion Detection

Abstract

Intrusion detection system (IDS) is crucial to security architecture of Internet of Things (IoT). In recent researches, the traditional machine learning and deep learning methods have been applied to the field of intrusion detection and achieved satisfactory performance. However, due to diverse IoT and dynamic network environment, it is difficult to use a single model for heterogeneous IoT networks and collect enough labeled data to train the new model. To solve these issues, we propose an intrusion detection approach based on heterogeneous transfer learning (HTL) for building an intrusion detection model with strong adaptability. Specifically, the approach consists of an Autoencoder architecture for aligning the heterogeneous features and lightweight Convolutional Neural Network (CNN) for unsupervised domain adaptation. Extensive experimental results on three public datasets reveal that the effectiveness of our proposed approach in the IoT environment with unlabeled and limited data.