Agile effort estimation using machine learning - A systematic review

This review paper examines the integration of machine learning techniques into Agile software development, primarily focusing on effort estimation. It evaluates existing methodologies for Effort Prediction (EP) in Agile Software Development (ASD) projects, emphasizing the Evolutionary Cost-Sensitive Deep Belief Network (ECS-DBN) model’s ability to predict task effort during the early stages of Agile projects. The model’s efficacy is assessed using realworld data from 160 tasks in Agile projects. Furthermore, the paper explores the applications of machine learning in various project management aspects within Scrum, such as sprint planning, backlog prioritization, and team performance prediction, as well as within Kanban, including workflow visualization, workload balancing, and lead time prediction. Emphasis is placed on the significance of data quality, algorithm selection, and the need for explainable AI. The paper concludes with a review of studies on software effort estimation in agile methodologies, highlighting the importance of machine learning algorithms in optimizing estimation formulas. Suggestions for future research include exploring additional metrics and applying machine learning techniques to industrial projects.