Improving story points estimation using ensemble machine learning

Abstract Agile software development (ASD) emphasizes iterative development, continuous feedback, and team collaboration, addressing the limitations of traditional methodologies. This research explores the application of machine learning (ML) to improve story point estimation in ASD, a critical practice for planning and prioritization. Traditional methods like Planning Poker often suffer from human biases and inconsistencies, leading to unreliable estimates. This study introduces an innovative ML-based ensemble stacking technique, combining RoBERTa, a transformer model for natural language processing, with BiLSTM, a neural network adept at handling sequential data. The research involves reviewing existing ML methodologies, developing the proposed model, and evaluating its effectiveness using 21,064 data points from 14 open-source projects. The model’s performance was assessed through Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Results show that the proposed ensemble model achieved lower MAE and MAPE, with performance improvements ranging from 4% to 32% over state-of-the-art models. While promising, the study suggests there is still room for further refinement, indicating the potential for ongoing advancements. This research contributes to the integration of ML in software engineering, offering a path toward more accurate and efficient project management.