Multi-label Classification of IoT Data Stream: A Survey

Abstract

The overall number of Internet of Things (IoT) devices is rapidly growing, generating a massive amount of continuous data stream. The data stream is arriving at a rapid speed, potentially unbounded, which has emerged due to smart services and advanced technologies. Data stream classification is a challenging task that must fulfil stream constraints such as limited memory, a single scan of data, and real-time response. In many emerging applications, stream instances could be associated with more than one class label, as when predicting a given movie genre, different labels may be given: action, horror, adventure, or all, and this refers to Multi-label Classification (MLC). This review mainly aims to review the literature on the multi-label classification task from 2014 to 2023. It examines state-of-the-art versatile MLC methods in general data streams and methods utilized for IoT applications, which are considered one of the main sources of data streams generated by IoT devices. It also focuses on two main challenges: class imbalance and concept drift. It encapsulates the well-known MLC tools and datasets utilized for this task. Moreover, it highlights the gaps that need further attention in future research.