An adaptive transformer model for anomaly detection in wireless sensor networks in real-time. (February 2023)

• Record Type: Journal Article
• Authors: Siva Kumar, A.; Raja, S.; Pritha, N.; Raviraj, Havaldar; Babitha Lincy, R.; Jency Rubia, J.
• Is Part Of: Measurement. Volume 25(2023)
• Publication Date: 2023-02
• Subjects: Anomaly detection -- WSN -- Sensor -- Transformer -- Spatio-temporal attention -- Adaptive; Detectors -- Periodicals; Measurement -- Periodicals; 530.7
• DOI: 10.1016/j.measen.2022.100625
• Languages: English
• ISSNs: 2665-9174

Abstract

Abstract: Defense, environmental monitoring, healthcare, home automation, and other fields are just a few of the many that make use of wireless sensor networks. There are sensor nodes in these networks that are sent out into the field to collect data and relay it back to the network's hub. The data streams may be infused with anomalous data due to interferences, malfunctioning nodes, etc. The security of the network depends on the prompt identification of any irregularities. Due to noise, missing data, and the dynamic nature of the network, identifying and detecting abnormal data is difficult.This study introduces a dynamic context-capturing deep learning model for sensor data anomaly detection, constructed as a Transformer with a spatio-temporal attention mechanism. The proposed model is trained and tested with a public dataset built from real-time sensor data captured in a water treatment plant. With an accuracy of 0.09142, F1 of 0.9109, and precision of 0.9191, the STA-Tran model outperforms the best existing methods.