GraphContrast IoT for self supervised graph based anomaly detection in multivariate industrial IoT sensor networks
Year: 2026
Type: article
Abstract: To ensure reliability and safety in industrial and cyber-physical systems, anomaly detection in multivariate IoT sensor data is essential. Most current autoencoder-based reconstruction models, recurrent architectures, and transformer-based approaches focus mainly on temporal patterns and do not sufficiently model interactions between different sensors, leading to lost contextual information and more false alarms. Many methods also rely on labelled anomalies and lack robustness to noisy or incomplete data. To address these problems, we propose a contrastive self-supervised graph neural network (GraphContrast-IoT) framework for multivariate time-series anomaly detection. The framework builds a hybrid graph that combines the physical topology and statistical correlation information and uses the proposed GCT-ADNet to learn the structural and temporal dependencies. Multi-view contrastive learning produces invariant, discriminative representations without anomaly labels, and embedding-based anomaly scoring increases anomaly-detection reliability. We estimate the hybrid adjacency matrix only from normal training data. It remains unchanged during training and inference, so the message-passing process is stable and efficient, and the graph topology is pre-defined. The F1-score for experiments performed on SWaT, WADI and SKAB were 0.941, 0.902 and 0.948, respectively. The corresponding precision values were 0.947, 0.910, and 0.953, while the recall values were 0.935, 0.894, and 0.943, respectively. The proposed framework outperformed the evaluated baseline methods and remained robust to perturbations in sensor data, including sensor noise and missing data, while providing low inference latency suitable for real-time monitoring. GraphContrast-IoT is therefore a scalable, label-efficient, robust and computation-efficient solution for industrial monitoring, cyber security and smart-infrastructure applications. The source code is publicly available at https://github.com/Sandya9876/GraphContrast-IoT . Clinical trial number: Not applicable.
Source: Discover Computing
Institution: Aditya University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,520
OpenAlex ID: W7225614603