Leveraging generative adversarial networks in project management for task estimation

Abstract

Accurate project estimation remains one of the most challenging aspects of project management, particularly when historical data is scarce, or project complexity is high. This study explores the use of Generative Adversarial Networks (GANs) to generate synthetic project data, including task duration estimates, by learning from historical data. Jira migration projects were used as an example. A new project with a distinct scope was introduced, and four GAN model variations were evaluated using Mean Absolute Error and Fréchet Distance to identify the most effective combination of parameters. The selected model was then used to generate task estimates for the new project, which were compared against traditional estimates based on the average values. Results show that the GAN-generated estimates were more conservative, highlighting potential hidden complexities and offering improved risk buffering. This approach demonstrates the potential of AI to enhance forecasting accuracy in project management and lays the foundation for future work on synthetic estimation of resources allocation, dependencies modeling, and risks identification.