SDLSC-TA: Subarea Division Learning Based Task Allocation in Sparse Mobile Crowdsensing

Sparse mobile crowdsensing (Sparse MCS), a new paradigm for large-scale fine-grained urban monitoring applications, collects sensing data from relatively few areas and infers data for uncovered areas. In Sparse MCS, the task allocation problem is simplified to the area selection problem since it is typically assumed that there were enough participants across the target sensing area. However, in many real scenarios, there is no guarantee the platform can find participants to execute tasks in vital areas. In this case, additional moving costs are incurred, which is not beneficial for the MCS platform as organizers are cost-sensitive. To address this problem, we propose a novel Subarea Division Learning based Task Allocation framework in Sparse mobile Crowdsensing (SDLSC-TA) that integrates subarea division learning, task allocation, and sensing map reconstruction. Different from existing research, we design the subarea division learning module to provide guidance for a more reasonable task allocation scheme. Specifically, subarea division learning utilizes the Iterative Self-organizing Data Analysis Techniques Algorithm (ISODATA) to perform uneven subarea division considering historical data and spatio-temporal correlations. Based on subarea division learning results, task allocation iteratively selects the most suitable cell and participant combining sensing levels, sensing, and moving costs. Finally, sensing map reconstruction utilizes Bayesian compressive sensing (BCS) to infer missing data while ensuring high quality. Using four typical urban sensing datasets, SDLSC-TA outperforms state-of-the-art sparse MCS frameworks by 15 percent lower total costs on average and 40 percent lower average sensing map error rate.