A Combined Framework for Agile Story Point Estimation Using AHP-Based Expert Weighting, Neutrosophic Z-Numbers, Bonferroni Aggregation, and XGBoost

Accurate story point estimation is crucial for effective agile planning in software development. Both primary methods have inherent limitations. Expert judgment is susceptible to subjectivity, whereas machine learning models require extensive, high-quality, project-specific datasets that are often unavailable. This study proposes a hybrid framework that emphasizes expert reasoning while incorporating data-driven reinforcement. Evaluators are weighted using the Analytic Hierarchy Process (AHP), and their item-level assessments are expressed as Neutrosophic Z-numbers (NZN) and aggregated using a Bonferroni Mean operator. The resulting estimate is then combined with a prediction from an Extreme Gradient Boosting (XGBoost) model. The framework was tested on 1500 backlog items from an agile team developing a Governance, Risk, and Compliance (GRC) product. “The classification performance, as indicated by an accuracy of 41.18%, a precision of 0.39, and a recall of 0.41, was moderate. In regression analysis, the Mean Absolute Error (MAE) was 1.05, the Mean Squared Error (MSE) was 3.15, and the Root Mean Squared Error (RMSE) was 1.77. Furthermore, the proximity to expert estimates was notable, with 78.88% of predictions falling within ±1 Fibonacci position and 94.65% within ±3 levels.” In terms of classification (accuracy 41.18%, precision 0.39, recall 0.41), regression (Mean Absolute Error (MAE) 1.05, Mean Squared Error (MSE) 3.15, Root Mean Squared Error (RMSE) 1.77, and proximity to expert estimates (78.88% within ±1 Fibonacci position and 94.65% within ±3 levels), the exact-class accuracy was moderate. Nonetheless, most predictions closely matched expert judgment. Significantly, the NZN–Bonferroni configuration achieved an average sprint completion rate of 97.41%, exceeding the fuzzy AHP benchmark of 93.94%. This framework provides agile teams with a transparent and repeatable method for improving the consistency of sprint planning.