Dynamic graph embedding‐based anomaly detection on internet of things time series

Abstract

Anomaly detection is critical in the internet of things (IoT) environment. To this issue, this study provides a novel approach for detecting anomalies in multivariate IoT time series. The proposed approach identified relationships between IoT time series to establish a dynamic graph and estimated the graph entropy to detect anomalies. The presented approach was applied to industrial IoT datasets. The results have shown that the presented method outperformed other models by 0.21 with respect to F1‐score. In addition, we used three distinct algorithms to detect the anomalies from the multivariate IoT time series. According to the results, the local outlier factor approach outperformed the others by 0.18 with respect to F1‐score.

Recent trends and advances in machine learning challenges and applications for industry 4.0

“…Jung and Li (2022) propose an entropy‐based dynamic graph embedding model using spurious correlation and Granger causality test to construct dynamic graphs of multi‐variate industrial control system cyber attack IoT time series, partitioned by an optimized time window size. Graph entropy detects anomalies between graphs grouped by entropy similarity in the embedding space.…”