Worker Recruitment with Distributionally Robust Optimization in Mobile Crowd Sensing

Abstract

Mobile crowd sensing (MCS) is an emerging sensing paradigm that leverages mobile devices for large-scale data collection, where an MCS platform recruits workers to perform time-sensitive sensing tasks. In practice, worker travel times are inherently uncertain due to dynamic urban environments—traffic congestion, road closures, and unplanned detours cause actual delays to deviate from historical patterns. However, existing worker recruitment methods typically optimize expected-case performance based on static historical distributions, providing no robustness guarantee under distribution shift. To address these challenges, this paper proposes DRO-IGR, a worker recruitment framework based on Distributionally Robust Optimization (DRO) and an Iterative Greedy Recruitment algorithm. The framework introduces a DRO Probability Estimation Engine that treats each worker’s true delay distribution as lying within a Wasserstein ambiguity set centered on its empirical history, with a radius that adapts to data sparsity, behavioral variability, and task importance. Strong duality reduces the resulting worst-case optimization to an efficient one-dimensional convex search. Driven by these robust probability estimates, a marginal-gain scoring rule iteratively selects worker–task pairs that maximize importance-weighted robust utility per unit cost. Extensive experiments on three synthetic spatial distributions and the real-world T-Drive Beijing taxi dataset show that, compared to existing approaches, DRO-IGR completes more tasks, achieves a higher total importance sum, and attains greater utility within the same budget constraint.