DAGAD: Dual Adversarial Learning Graph Anomaly Detection in Multivariate Time Series Data

The increasing number of high-dimensional time series data poses challenges for traditional anomaly detection methods that rely on supervised approaches. In this paper, we propose a stable and novel method called dual adversarial learning graph anomaly detection, which effectively captures complex data relationships and accurately detects anomalies away from them. Our framework utilizes a graph structure to capture complex relationships between variables, while the dual adversarial training overcomes the inherent limitations of autoencoders. In addition, we incorporate the prediction techniques to enhance the ability to identify anomalies. We evaluate our proposed model on publicly available real-world datasets and compare its performance against various existing methods. The experimental results demonstrate that our method achieves more accurate anomaly detection compared to baseline methods.