An IoT-Based Anomaly Detection and Identification Approach for Gas Sensor Networks

Environmental gas sensors are important tools towards the assessment of air quality and the quantification of the presence of polluting gases. Especially when arranged in gas sensor networks, such devices can provide relevant information on the distribution of pollutants in affected areas, such as in urban environments. In such a distributed measurement setup, the continuous maintenance of the measurement devices is a particularly important issue, since low-cost devices are prone to different types of faults or interfering environmental factors, which can have a negative impact on the sensing quality of the network. In this paper, we propose a machine learning-based anomaly detection and identification approach for environmental gas sensor networks, which operates efficiently on-the-edge by partially distributing the algorithm across the neighboring nodes. Our investigations furthermore included the introduction of a simulation use-case for the proper assessment of such a connected anomaly detection scenario, which is adjusted towards gas sensor-related anomalies. Evaluating the proposed method on this simulation scenario, the approach shows convincing accuracy levels for the detection and the identification of moderate sensor anomalies, while even performing considerably well for more subtle anomalies.