Unsupervised Federated Learning based IoT Intrusion Detection

Krishna Yadav, Brij B. Gupta, C Hsu, Kwok Tai Chui

20212021 IEEE 10th Global Conference on Consumer Electronics (GCCE)26 citationsDOI

Abstract

Machine learning has been widely used these days to detect novel intrusions across IoT devices. Supervised-based machine learning techniques need labelled datasets to train a model. Due to privacy reasons, these days, people don’t share the dataset generated across their devices with external authority. When datasets are not aggregated centrally, it becomes very difficult to process the unlabelled data and train a model across edge devices. Considering these drawbacks, we have brought an unsupervised deep learning approach that uses autoencoders to learn from unlabeled data. Our approach uses federated machine learning and can be trained across the unlabeled dataset of edge devices without compromising people’s privacy. We have tested our approach against CICIDS 2017 dataset in a federated environment and have got an accuracy of 97.75% in detecting intrusions.

Topics & Concepts

Computer scienceArtificial intelligenceMachine learningUnsupervised learningEnhanced Data Rates for GSM EvolutionProcess (computing)Intrusion detection systemFederated learningDeep learningInternet of ThingsLabeled dataEdge deviceEdge computingSupervised learningData miningArtificial neural networkCloud computingWorld Wide WebOperating systemNetwork Security and Intrusion DetectionInternet Traffic Analysis and Secure E-votingAdvanced Malware Detection Techniques