JoWUA

Abstract

Accurate effort estimation and prior risk assessment in the context of distributed software project pipelines managed via asynchronous workflows, where agile teams separately located across the globe work under changing requirements, is a challenging problem due to multifaceted uncertainties. This work introduces a comprehensive framework that incorporates adaptive machine learning to improve real-time estimation and dynamically identify critical high-risk tasks. The predictive algorithms are tailored to changing project data using ensemble learning and online adaptation methods which effectively tackle concept drift and cross-contextual variability concerns. The empirical study was conducted on datasets obtained from agile distributed development settings involving several teams working across different time zones. The findings reveal that estimation accuracy was notably enhanced, achieving up to 27% reduction in RMSE compared to static models. Furthermore, the adaptive risk classifier outperformed other scenarios with high F1 scores, particularly in identifying components that were likely to be delayed and defect-ridden. With its scalable low-latency inference, the framework is positioned to be incorporated within CI/CD systems. This research establishes a rich intelligence for planning frameworks in software projects, thereby improving automated frameworks under the distributed multi-software engineer systems environment, and sophisticated, sharp decision-making algorithms rely on accurate real-time data.