Adaptive Window Based Sampling on The Edge for Internet of Things Data Streams
Taimur Hafeez, Gavin McArdle, Lina Xu · 2020
The Internet of Things (IoT) generates a massive amount of data. To relieve the pressure placed on cloud and network services, sampling techniques have been applied to remove the redundant data. Thanks to the recently emerged Edge Computing (EC) paradigm, sampling on the edge rather than on the IoT end devices can be more intelligent. Most existing sampling algorithms, applicable to edge, do not consider the complex use cases such as predictive analytics. In this paper, we propose a real-time Adaptive Window Based Sampling (AWBS) algorithm to dynamically sample IoT time-series data on the edge. We have verified AWBS on an IoT machine health dataset from NASA using Local Outlier Factor, a well-known unsupervised abnormality detection method. The results were compared with two state-of-the-art sampling algorithms. The experimental results show that AWBS outperforms the other two algorithms by 1) effectively reducing the data to 6.91% and 2) accurately identifying the abnormal behaviour of the machine.