Machine Learning Based Software Effort Estimation In Agile Based Projects By Using Scrumban Methodologies

Abstract

Effort estimation in software development is important in the context of an Agile software development process due to the changing requirements of the customers and the dynamics of the business environment, which leads to uncertainties in the planning and resource management process. The conventional techniques of effort estimation do not yield accurate predictions. This study proposes a novel Multi Learning-Scrumban Framework, a Machine Learning-based Software Effort Estimation model integrated with Scrumban methodologies. The proposed framework combines Scrum-based sprint planning and backlog management with Kanban workflow visualization and monitoring to enhance estimation accuracy and project adaptability. Historical Agile project datasets containing Story Points, Sprint Velocity, Team Experience, Project Complexity, Backlog Size, Sprint Duration, and Requirement Volatility were utilized for model development. Various advanced machine learning models such as Random Forest, Support Vector Regression, and Gradient Boosting Regression were developed and tuned to estimate the software effort in person-hours. The experimental assessment carried out for more than 1,500 cases of Agile projects showed the efficiency of the proposed ML-SCRUMBAN Framework. The framework provided 96.82% estimation accuracy, 96.31% precision, 95.88% recall and 96.09% F1-score. Additionally, it decreased Mean Absolute Error by 38.9%, Root Mean Square Error by 37.3% and increased PRED(25) to 98.74% compared to traditional methods of Agile estimation. Thus, it is evident that the proposed framework substantially improves estimation accuracy, sprint planning efficiency and the general project performance in Agile software development environment.