MAB-RP: A Multi-Armed Bandit based workers selection scheme for accurate data collection in crowdsensing

Abstract

Accurate data collection from workers is crucial for the success of Mobile Crowd Sensing (MCS) applications. However, Current studies exhibit several drawbacks. Firstly, the workers' sensing qualities remain unknown even after the platform acquires the data submitted by the workers, known as the Post Unknown Worker Selection (PUWS) problem. Secondly, systematic deviations between worker data and the Ground Truth Data (GTD) reduce the quality of MCS applications. Thirdly, the data collected by workers for different tasks may vary in accuracy, resulting in low-quality data collection. To address these challenges, we propose a novel Multi-armit-based worker selection scheme with reputation and preference (MAB-RP). The proposed scheme aims to select credible workers for high-quality data collection through trust identification, thus addressing the PUWS issue after worker recruitment. Additionally, the scheme employs a learning-based approach to identify and correct the gaps between the sensed data and the GTD, ultimately improving the accuracy of data collection. Lastly, a matching-based approach is used to identify workers' sensing qualities for different tasks, further enhancing the accuracy of data collection in MCS. Extensive simulations on real-world datasets demonstrate that the proposed MAB-RP scheme outperforms previous strategies in terms of both data quality and cost.