Fisher Information-Empowered Sensing Quality Quantification for Crowdsensing Networks
Xiang Chaocan,Fan Xiaochen,Chen Chao,Gong Liangyi,Guo Songtao
DOI: https://doi.org/10.1007/s00521-020-05501-6
2021-01-01
Neural Computing and Applications
Abstract:The sensing quality is critical for crowdsensing networks. However, it is still challenging to achieve reliable and accurate quantification on sensing quality. As most existing methods only quantify individual-level sensing quality, they cannot be simply extended to measure network-level sensing quality. Meanwhile, there are two critical challenges for the non-trivial quantifications of network-level sensing quality. First, it is quite daunting to accurately measure the sensing quality, due to the uncertainties in crowdsensing, such as random noise in sensing data and complicated inference of information. Second, it is incredibly difficult to conduct repeatable experiments for multiple times, with the natural dynamics of crowdsensing networks as well as the uncontrollable behaviors of crowdsensing participants. To address the above challenges, in this work, we devise a novel quantification metric to measure the uncertain sensing quality of crowdsensing networks. The proposed metric is based on the confidence interval, exploiting the asymptotic normality property of unbiased estimations. We further leverage the Fisher information in crowdsensing data and successfully derive the confidence interval without redundant multiple computations on repeated experiments. The trace-driven evaluations demonstrate that the proposed method can achieve remarkably accurate quantifications on the network-level sensing quality, outperforming the existing works.