An Imperfect Monitoring Game-Theoretic Approach for Crowdsourcing Reliable Wireless Data

Gathering useful and trustworthy wireless data using crowdsourcing to train/validate machine learning (ML) algorithms can be difficult due to two factors: 1) correctness and reliability of the gathered data from various independent wireless access points (APs) can be unknown, and 2) designing a metric that can capture the value of the collected data for the considered model can be a challenge. To address the challenge of reliable data collection, we propose a game-theoretic crowdsourcing-based wireless data collection mechanism that can be used to reliably collect the data. We also propose a metric that can capture the value of the collected data for the considered model. Our proposed method provides protection against selfish deviations and unlike other game-theoretic works does not make an assumption that the actions of the crowdsourcing agent APs are known to the crowdsourcing entity. We consider a technique that can be used to infer the actions of agent APs and propose that the participants utilize the n-round Win-stay lose shift (WSLS) strategy. In our work, we compare the performance of the proposed strategy against various other game-theoretic strategies, such as always high, always low, WSLS with probability p, and tit for tat. In our performance evaluation, we utilize both real and synthetic wireless channel utilization data. Our results show that the n-round WSLS strategy outperforms the other game-theoretic strategies.