Integrating Synthetic Data Modelling into an Adaptive Sampling Framework for IoT Devices

Wireless Sensor Networks (WSNs) often rely on constrained devices with limited processor and power resources. Therefore, many approaches in the literature have studied data reduction methods to relieve these resources. Static scheduling algorithms are state of the art, but novel adaptive scheduling methods gained momentum in recent years as well. In the study presented in this paper, we combine enhanced adaptive scheduling algorithms with synthetic data, created using a generative machine learning model. This allows us to replace some of the sensed data points by generated ones. In consequence, we can increase the time interval between collecting two samples, which alleviates the power resources of the sensors significantly. Utilizing working environment noise data sensed in Trondheim, Norway, we analyzed how generative machine learning models can enhance data collection in resource-constrained wireless sensor networks. The results of this analysis reveal that using synthetic data to reduce the sampling rate by over 50% greatly improves data reduction efficiency, outperforming traditional adaptive sampling frameworks. To the best of our knowledge, our approach constitutes the first effort to leverage generative models for crafting adaptive sensing policies in WSNs.