Research on user recruitment algorithms based on user trajectory prediction with sparse mobile crowd sensing.

Area of Science:

• Computer Science
• Data Science
• Artificial Intelligence

Background:

• Sparse mobile crowd sensing reduces costs by collecting data from a subset of users and inferring the rest.
• Participant trajectories yield data of varying value, necessitating effective user selection for high-value data collection.
• Accurate user trajectory prediction and recruitment are crucial for optimizing sparse mobile crowd sensing.

Purpose of the Study:

• To improve user trajectory prediction accuracy in mobile crowd sensing.
• To develop an effective user recruitment algorithm for maximizing data value under budget constraints.
• To enhance the overall accuracy of data inference in sparse mobile crowd sensing.

Main Methods:

• Proposed STGCN-GRU algorithm combining Spatio-Temporal Graph Convolutional Networks (STGCN) and Gated Recurrent Unit (GRU) for trajectory prediction.
• Developed an Action Deep Q-Network (ADQN) algorithm, a reinforcement learning approach to improve user recruitment by addressing Q-network overestimation.
• Evaluated algorithms using standard metrics (FDE, ADE) and real-world datasets.

Main Results:

• The STGCN-GRU algorithm demonstrated superior performance in trajectory prediction accuracy compared to existing methods.
• The ADQN algorithm effectively improved user recruitment accuracy, leading to better data inference under budget limitations.
• Experimental results validated the practical effectiveness of both proposed algorithms on real datasets.

Conclusions:

• The STGCN-GRU and ADQN algorithms offer significant improvements in user trajectory prediction and recruitment for sparse mobile crowd sensing.
• These advancements enable more accurate and cost-effective data inference, optimizing resource allocation in sensing networks.
• The proposed methods provide a robust framework for enhancing the efficiency and effectiveness of mobile crowd sensing applications.