A Comprehensive Evaluation of Generating a Mobile Traffic Data Scheme without a Coarse-Grained Process Using CSR-GAN.

Area of Science:

• Data Science
• Computer Science
• Network Engineering

Background:

• Mobile traffic data analysis is crucial for urban planning, but escalating data volumes increase storage costs.
• Existing methods using generative adversarial networks (GANs) reduce data but still require storing coarse-grained representations.
• This necessitates more efficient data generation and storage solutions for mobile traffic data.

Purpose of the Study:

• To propose a novel scheme for generating mobile traffic data using conditional-super-resolution GAN (CSR-GAN).
• To eliminate the need for a coarse-grained data representation process, thereby reducing storage requirements.
• To evaluate the accuracy and storage efficiency of the proposed CSR-GAN method compared to traditional approaches.

Main Methods:

• Implementation of a conditional-super-resolution GAN (CSR-GAN) model for mobile traffic data generation.
• Utilizing real-world mobile traffic datasets for experimental validation.
• Comparative analysis of storage costs and data generation accuracy against existing methods.
• Investigating the impact of CSR-GAN architecture variations on performance and storage needs.

Main Results:

• The proposed CSR-GAN scheme reduces storage costs by up to 45% compared to traditional methods.
• The generated mobile traffic data achieves an accuracy of 94% in reconstructing original data.
• Experiments revealed an optimal relationship between traffic data volume and CSR-GAN model size.

Conclusions:

• CSR-GAN offers a significant reduction in storage costs for mobile traffic data management.
• The method effectively generates high-accuracy mobile traffic data without intermediate coarse-grained representations.
• Findings suggest that CSR-GAN is a viable and efficient solution for large-scale mobile traffic data analysis and planning.