Abstract The extensive use of Internet of Things devices leads to the generation of vast amounts of high-frequency time-series data. Analysing these data provides valuable insights for business decision-making. However, to provide timely decisions, we should be able to analyse time-series streaming data on the fly. While numerous advanced analytical techniques have been developed to rapidly identify behaviour profiles that highlight prominent patterns and enable anomaly detection in time-series data, these methods are mainly tailored to static datasets rather than dynamic data streams. In addition, existing methods for incrementally detecting and visualising patterns and anomalies in multi-variate time-series streaming data still have limitations. This paper proposes a swarm intelligence-based visual analytics approach and algorithm to learn behaviour profiles incrementally, automatically determining the appropriate number of profiles from the data stream. Each profile is a network of data prototypes representing the data distributions. A new visualisation method is developed to visualise behaviour profiles that can reveal cyclic patterns and anomalies without requiring a dimensionality reduction step. Experiments with real-world time-series datasets show that the proposed approach can capture, cluster and visualise key patterns and anomalies, providing actionable insights from the data.
Pham et al. (Mon,) studied this question.