IoT security: a systematic literature review of feature selection methods for machine learning-based attack classification

Abstract

In the age of the internet of things (IoT), ensuring security is crucial to protect the interconnected devices and systems. The capacity to identify cyberattacks is essential for IoT security, hence many academics have focused their efforts on developing powerful classification models that can identify intrusions to protect IoT infrastructure. One key factor in creating successful classification models for IoT security is feature selection. To assist researchers and practitioners in selecting the appropriate feature selection methods, this paper presents a systematic literature review of the literature on feature selection approaches for machine learning-based attack classification models in IoT security using IoT datasets. By analysing data from 1,272 studies published between January 2018 and December 2022 using preferred reporting items for systematic literature reviews and meta-analyses (PRISMA) guidelines, the review identifies 63 primary studies that meet inclusion criteria. The primary studies are analysed and categorised to answer research questions related to current practices, feature selection methods, benchmark IoT datasets, feature selection validation methods, limitations, challenges, and future directions. The review provides valuable insights for researchers and practitioners seeking to incorporate effective feature selection approaches in IoT security.